System and Method for Location-based Construction Scheduling

Field of invention

The present invention relates to a method and a system for scheduling a construction process, in particular using location-based logic.

Prior art

Planning of construction processes, for example building construction processes, is challenging, especially when the construction area is large and complex. Substantial delay of milestones occur frequently. Besides, the actual cost of the works often exceeds the agreed price.

In practice, when a time schedule is provided by using a prior art solu- tion based on all information available, the schedule is easily outdated due to changes that effect the expected remaining construction time and the total cost. Accordingly, in a large and complex construction pro- cess, the total cost often is increased because of unexpected chal- lenges, and dealing with such challenges can be time consuming and expensive. Moreover, it is normally difficult, if not impossible, to keep the time schedule updated due to unforeseen challenges that come up. In real life, few of the prior art solutions are configured to receive and take into account minor and major changes that effect the construction process. Furthermore, a number of construction specific dependencies are not taken into consideration when using many prior art solutions. Therefore, the prior art solutions typically fail to provide realistic con- struction schedule. Consequently, the expected remaining construction time and the total cost cannot be predicted in an accurate manner.

Accordingly, it would be an advantage to have a method and a system for providing construction schedules by which expected remaining con- struction time and total cost can be anticipates more realistically. For this objective, the prior art does disclose computerised methods that assist in the construction scheduling. An example is disclosed in US2019/0205484, where the system uses construction recipes and relationships between the construction ele- ments. A construction recipe comprises a formula for building a con- struction element and comprises multiple construction operations inter- related by logical relationships. The same construction recipe is then as- signed to construction elements of the same type. An operation is de- fined by required resources drawn from a resource pool.

In this approach, the units from which relationships are created are the construction element. The relationships between the various construc- tion elements represent the entire construction. This is in its basic ap- proach fundamentally different from location-based planning, which takes offset in sub-locations of the entire construction and seeks repli- cation of construction sequences at various sub-locations. For example, if a building has largely identical offices, a construction sequence for a single office can be repeated for as many similar offices in the building, and similarly for construction sequences related to flooring, wall setup and ceiling. For the construction method in this location-based ap- proach, the total workflow for the construction is simplified and by far less complex than in US2019/0205484, as a relatively simple construc- tion sequence is merely repeated in location-based construction sched- uling without the need of the complexity of all relations between reci- pes. The number of logic dependencies in location-based construction planning is much reduced in comparison to the approach in US2019/0205484.

An example of a computerised location-based scheduling system is dis- closed in US2010/043342 by Seppanen et al. As explained therein, and in line with the above arguments, for simplification, location-based con- struction schedulings systems use a location-based logic where it is as- sumed that for multiple locations, the logic for specific construction se- quences is similar, so that specific construction sequences can be repli- cated from one location to another. The location-based approach can af- ford having included several layers of logic, including different risk factors and internal as well as external dependency systems without compromising overall accuracy due to increased complexity. Despite such multilayer logic, the model is relatively robust.

It is emphasized that this multilayer approach is in contrast to the sys- tem as disclosed in US2019/0205484, in which only a single layer of logic is used. The latter is due to the complexity of the recipe-based cal- culation without replication of construction-sequences, where, instead, each construction element has a dependency on other construction ele- ments, which requires a very high number of recipe relationships be- tween associated elements, making the model and calculations complex already in a single layer of logic.

Whereas in location-based construction modelling, a construction model can be copied from one location to another for the same task, this is not the case in the approach of US2019/0205484, where a complete and complex relationship model between the construction elements is re- quired, irrespective of the fact that multiple construction elements and recipes can be identical.

This advantage of location-based scheduling over other systems, in par- ticular the system disclosed in US2019/0205484, implies that multiple layers of construction scheduling can be used, including not only inter- -al but also external relationships between activities and locations, without compromising calculation efficiency by computation to an extent where results become unreliable.

Despite the strength and advantageous approach in US2010/043342, the method implies a few disadvantages. For example, uncertainties are included as risk factors. Although, such risk factors appear reasonable and as an apparent improvement of the overall computerised method, it introduces uncertainties in the calculation which have to be carefully weighted in order not to calculate in directions which do not lead to use- ful and reliable results. Further, although, US2010/043342 describes that the method as disclosed does a comparison between different cal- culations, there is not disclosed any concrete specific way of calculation for optimising the model with respect to reliability of the best way for- ward when changes occur during the construction. In particular the model as disclosed in US2010/043342 suffers from focusing on needs for machinery and staff rather on the actually available resources and changes therein.

It would therefore be desirable to provide an improvement in computer- ised location-based construction scheduling.

Summary of the invention

It is an objective of the present invention to provide an improvement in the art. In particular it is an objective to provide an improved method for computerised location-based construction scheduling, especially by increasing efficacy and reliability in the calculation. This and further ob- jectives and advantages are achieved by a method and system as de- fined in the independent and dependent claims and explained in detail in the following description and exemplified by illustrations in the ac- companying drawings.

The method is a method for location-based construction scheduling, wherein the method comprises

- providing a two-dimensional or three-dimensional model of a con- struction object;

- dividing the construction object into sub-locations and assigning pre- determined construction elements and construction sequences to each sub-location, wherein the division into sub-locations is based on criteria of replication of the predetermined construction elements and construc- tion sequences from one to another of at least two of the sub locations;

- assigning required resources and durations of the construction se- quences to the sub-locations;

- providing resource data about available resources as input for the con- struction scheduling, the resources data comprising data on available workforce and their productivity rates, available construction machin- ery, and available construction materials;

- providing constraints on time-dependent and logical-dependent orders of the construction sequences;

- by a computer that is programmed with a computer program perform- ing an iterative computer-implemented calculation process with multiple iterations, the process comprising automatic variation of selected con- struction parameters and including sizing, splitting, and allocation of workforce between sub-locations, and calculating for each iteration a construction schedule for construction of the construction object in de- pendence of the construction parameters, each iteration including distri- bution of the available resources for the construction sequences with the construction elements, taking into regard the constraints, and calcu- lating for each iteration the requirements for time and costs of the re- spective construction schedule,

- according to predetermined criteria providing one of the calculated constructions plans as optimum by the for pursuing the construction with this construction schedule.

By providing a two-dimensional or three-dimensional model of a con- struction object it is possible to achieve the technical effect of establish- ing an overview of the entire construction object. This step enables a basis for carrying out several additional process step in order to achieve further useful technical effects.

By dividing the construction object into sub-locations and assigning pre- determined construction elements and construction sequences to each sub-location, wherein the division into sub-locations is based on criteria of replication of the predetermined construction elements and construc- tion sequences from one to another of at least two of the sub locations is possible to achieve the technical effect of establishing a more detailed overview of sub-locations, their predetermined construction elements and construction sequences. By introducing the construction sequences is possible to achieve the technical effect of establishing a link between the construction elements and the construction sequences. Hereby, the method is capable of effectively take into consideration the effect of any link between the construction elements and the construction sequences.

By assigning required resources and durations of the construction se- quences to the sub-locations it is possible to achieve the technical effect of enabling a real time simulation of the effect of maintaining or chang- ing (the quantity of) resources. This is essential since it enables the method to simulate and thus anticipate the effect of any change in the quantity of resources.

By providing resource data about available resources as input for the construction scheduling, the resources data comprising data on availa- ble workforce and their productivity rates, available construction ma- chinery, and available construction materials it is possible to achieve the technical effect of making it possible to provide real time simula- tions based on the actual available resources via the resource data.

By providing resources data comprising data on available workforce and their productivity rates, available construction machinery it is possible to achieve the technical effect of calculating constructions plans based on: a) the actual resources data and b) resources data that deviant from the actual resources data.

By providing constraints on time-dependent and logical-dependent or- ders of the construction sequences it is possible to achieve the technical effect of providing a realistic simulation that takes into consideration time-dependent and logical-dependent orders of the construction se- quences.

The method applies the step of by a computer that is programmed with a computer program performing an iterative computer-implemented calculation process with multiple iterations, the process comprising automatic variation of selected construction parameters and including sizing, splitting, and allocation of workforce between sub-locations, and calculating for each iteration a construction schedule for construction of the construction object in dependence of the construction parameters, each iteration including distribution of the available resources for the construction sequences with the construction elements, taking into re- gard the constraints, and calculating for each iteration the requirements for time and costs of the respective construction schedule;

- according to predetermined criteria providing one of the calculated constructions plans as optimum by the for pursuing the construction with this construction schedule.

Hereby, it is possible to achieve the technical effect of establishing a simulation that in an effective manner takes into account the constrains of the entire construction object.

In particular, it has been found useful to use artificial intelligence (Al) and an iterative process in the computerised construction scheduling. Furthermore, during the construction, the computerised scheduling model is periodically repeated in order to take into account changes during the construction, for example delays and change of available materials, staff, machinery, and equipment. By periodically updating data records of the available pool of resources, for example weekly or daily, and using these data as input for renewed calculations as well as the actual status of the construction, realistic up-to-date results are achieved by the periodically repeated calculations.

The method and system as described herein are capable of providing realistic construction schedules, so that the expected remaining con- struction time and the total cost can be anticipated more realistically. Also achieved is an improved way of accessing data of relevance to carry on an optimal management of the construction process. The method and system provide realistic construction schedules in which the total cost and the duration of the construction period can be predicated accurately at any point of time of the construction process if calcula- tions are periodically repeated.

In particular, a computer-implemented method is provided for location- based construction scheduling. Furthermore, a construction scheduling system is provided comprising a computer programmed to execute the method. Also provided is a computer program product comprising in- structions to cause the system to execute the steps of the computer im- plemented method as explained herein. Additionally, the objective is achieved with a computer-readable medium having stored thereon the computer program.

For the method, typically, three-dimensional models of a construction object are provided, although, in some cases, a two-dimensional model is sufficient. Non-limiting examples of construction objects are buildings or groups of buildings, for example office buildings or residential build- ings, roads, railroads, bridges, tunnels, factories, and energy plants, such as wind or solar power plants.

For the computer calculation of the construction schedule, the construc- tion object is divided into sub-locations to each of which predetermined construction elements and construction sequences are assigned, includ- ing required resources and durations of the construction sequences.

This can be done automatically by the computer program, or it is part of the input data to the computer.

As location-based scheduling has an advantage of replication, the divi- sion into sub-locations is based on criteria of replication of the predeter- mined construction elements and construction sequences from one to another of at least two of the sub locations. As an example, for an office building with a plurality of similar offices, a division into levels and of- fices as sub-locations would be useful. For a residential building, the stairways and apartments would be useful sub-locations, especially if the apartments are alike. In detailed schedules, the locations can be further sub-divided into ever smaller sub-locations such as rooms.

Typically, the sub-locations into which the construction elements are di- vided have a size of less than 1000 m ² , advantageously under 300 m ² to 400 m ² . In buildings with a fast-paced construction process typically less than 100 m ² . Generally, the size is larger than 10 m ² , potentially larger than 20 m ² .

For a wind power plant, separation into sub-locations with a single wind turbine would be useful. A construction sequence would then involve earth work, foundation, assembly of tower, cable trays, installation of cabling, nacelle, rotor and wings mount, testing in this order.

Additionally, there are provided constraints on time-dependent and logi- cal-dependent orders of the construction sequences. This is necessary, for example, because ceiling cannot be constructed before there is a support therefore. Or a rotor cannot be mounted before a wind turbine tower is erected.

In order to achieve a realistic construction schedule, there are provided input resource data about available resources, including but not delim- ited to available workforce and their productivity rates, available con- struction machinery, for example cranes and lorries, and available con- struction materials.

The computer implemented method is in an iterative computer-pro- grammed calculation process with multiple iterations, typically in the range of 100-1000 iterations, for example 100-300 iterations. However, in contrast to prior art Monte Carlo iterations, the invention uses a method of varying selected construction parameters from one iteration to another. Useful construction parameters are sizing, splitting, and al- location of workforce between the sub-locations, CO2 emission, cost per hour, mobilization cost, as well as allocation of machinery, for example cranes, and transport logic of construction material. The program then calculates for each iteration a construction schedule for construction of the construction elements in dependence of the construction parame- ters, with each iteration using an amended set of construction parame- ters, and each iteration including distribution of the available resources to the construction elements and the construction sequences, taking into regard the constraints. Each iterative calculation has as output numbers for the requirements for time and costs of the respective con- struction schedule. After the various iterations, a comparison is done, and an optimum plan is provided, for example selected, from the plans of the iteration, according to predetermined criteria, typically cost and/or time, and the construction is pursued according to this construc- tion schedule.

Advantageously, artificial intelligence, Al, is used on the process. In some of such embodiments, the method comprises assigning an evalua- tion parameter to each iteration, wherein the evaluation parameters re- flect whether an iteration yields an improvement or not as compared to the foregoing iteration or iterations, especially whether the iteration yielded a decrease or an increase in costs and/or time. Such evaluation parameter can then be used as a criterion for the provision, for example selection, of the optimum plan.

It has been found useful to create an evaluation parameter mapping, where the mapping is reflecting a directions and/or local areas for opti- mized construction parameters, typically in a direction of minimization of costs and/or time, depending on the direction of variation of the se- lected construction parameters. Such evaluation parameters are then used in a programmed artificial intelligence, Al, machine learning rou- tine for determining directions for further iterations with the criteria of optimizing the selected minimization of costs and/or time of the con- struction. Once a direction has been determined, the program is contin- uing the iterations along the direction(s) as determined by the Al ma- chine learning routine until a predetermined number of iterations has been reached or until a useful local minimum or further local minima for costs and/or time are found.

In order for the construction to be optimised during the construction process, it is advantageous to periodically updating status data on ac- tual status of the construction and re-running the iterative computer- programmed calculation process on the basis of the updated status, and providing a new construction schedule for pursuing the remaining con- struction, the new construction schedule being optimized on the basis of the status data relatively to the previous construction schedule.

As a further advantageous measure, the method comprises periodically updating the resource data and re-running the iterative computer-pro- grammed calculation process on the basis of the updated resource data, and providing a new construction schedule for pursuing the construc- tion, the new construction schedule being optimized on the basis of the updates resource data relatively to the previous construction schedule. For example, for the optimization, the resources are redistributed be- tween sub-locations.

In further advantageous embodiments, the method comprises storing additional evaluation parameters into the evaluation parameter map with each subsequent re-running of the iterative computer-programmed calculation process after update of the resource data and construction status data. These earlier stored evaluation parameters can be used by the Al learning routine for later calculations. In particular, for each fur- ther calculation process, evaluation parameters from earlier iterative calculation processes are included for the determination of directions for further iterations by the Al machine learning routine in order to obtain an accelerated re-calculation of minimization of costs and/or time in de- pendence of the actual available resources and the actual construction status.

This is in particularly useful due to the fact of the location-based sched- uling where even a small improvement of one sub-location may yield a substantial cost and/or time saving due to the replication of the con- struction sequences to other sub-locations.

In location-based construction scheduling, various layers of logic are implemented, as also discussed in US2010/043342, for example a five- layer logic. It has been found useful to implement such five layers in the computer implemented invention herein.

Advantageous, the method comprises providing the constraints in the computer implemented method as multiple layer logic constraints, in- cluding at least one of the following, for example all of the following:

- logical and timely relationships between activities within a sub-loca- tion, wherein a sequential relationship applied between two tasks is ap- plied in each sub-location where both tasks exist, the two task being part of a construction sequence; for example, a gypsum board wall must be built before it can be painted in a sub-location.

- logical and timely relationships driven by hierarchy levels of location breakdown structure, determining timely relationships between different construction sequences at -different sub-locations; for example, con- struction of the foundation before the lower part of the wind turbine is erected;

- logical and timely relationships between sub-locations within predeter- mined tasks, including movement of workforce from, to and through lo- cations; for example the workforce for the activity "electrical installa- tions" is moving from stairway 1 ground floor apartments and up through the building, and only thereafter to stairway 2, and again up through the building;

- logical and timely relationships between construction sequences of dif- ferent sub-locations including location lags (activity delay at one loca- tion depending on activities on a different location) of portions of con- struction sequences, which are necessary to be finished prior to starting a different construction sequence at a different sub-location; for exam- ple, two levels of the wind turbine tower must be installed before the electrical installations can begin at the sub-location at ground level; - logical and timely relationships between any tasks at different loca- tions, including linking different construction phases together; for ex- ample, phase-wise commissioning of batches of wind turbines inside a wind turbine power plant requires finishing of the respective number of wind turbines before commissioning can be done.

For example, the computer system for the method comprises various computation modules. A structure module comprises information about the components forming construction elements of the sub-locations, for example building structures, such as walls. A calculation module is con- figured to calculate activity durations of construction sequences with lo- cation-based construction logic and hereby provide a time schedule. An optimization module configured to calculate a time schedule based on the provided resource data and further information on requirements for the construction, wherein activity durations are calculated in a location- based dependency, wherein the optimization module is configured to calculate a time schedule in which a number of construction specific de- pendencies and constraints are taken into consideration.

Optionally, for the provision of resource data, a database is configured to receive and store information uploaded by a number of providers. Apart from data on work force, construction machinery, and materials, it also contains information including one or more of the following: CO2 emission, cost data, productivity factor data, mobilization costs, hourly costs, and production rates data.

Hereby, a system is provided that yields realistic construction schedules for managing a construction process with reliable prediction of cost and duration at any point of time during the construction process.

As mentioned above, for the construction, a three-dimensional model of a construction object is provided. The model contains information about position and geometry of the components of the construction and their locations. The positions are defined with respect to a predefined refer- ence coordinate system. For example, the centre of mass of a structure is defined with respect to the reference coordinate system. Optionally, corner point of each structure is defined with respect to the reference coordinate system. The construction model also contains information about the orientation of the structures at the sub-locations. The orienta- -ion is defined with respect to the predefined reference coordinate sys- tem. As all structures are well-defined with respect to position, geome- try and orientation, the system and method according to the invention can calculate and simulate the schedule on basis of the exact number of each type of structures in each sub-location. Non-limiting examples of such structures are concrete foundations, wind turbine towers, asphalt sections, road tiles, doors, walls, floors, ceilings, gypsum boards and even smaller structures, such as cable trays, equipment, reinforcement rods, and power outlets. Typically, the parameters, including geome- tries, for the structures used in the constructions are received from ex- ternal systems and used in the calculations.

Advantageously, the geometries are contained in a model module that gives necessary geometrical information to an optimization module.

Since the geometries of the structures are defined as well as the overall construction, the system and method can handle the volume of a struc- ture with precision. Moreover, the system and method can handle the volume or length of that portion of a structure that is located in any part of the sub-locations of the construction complex. Thus, the system and method can calculate the price, for example based on known cost per volume or length. The system and method can also calculate the es- timated duration of a task, based on known working capacity of provid- ers such as the time specific work rate cost. The system and method are capable of calculating the estimated cost for one or more tasks to be carried out, or have been accomplished, for any selection made up by one or more locations.

In some embodiments, the database used for the calculations is config- ured to receive and store information uploaded by a number of different providers, especially with respect to resources. Resource data include data about staff, for example bricklayers, carpenters, electrical techni- cians or other craftsmen. To the extent needed, the resource data also includes data about construction material such as concrete, reinforce- ment structures, flooring, windows, doors, electrical installations, con- crete foundations, scaffolding, asphalt, road tiles, and roofing. Resource data also include cost data about parts and labour, such as cost of con- struction material and time specific pay/salary. Cost data may include cost for transportation, and optionally costs for rented machinery or tools.

Productivity factor data may define the efficiency of manpower in de- pendency of the number of persons or in dependency of the type of work to be carried out. This is important, as a doubling of the staff may not lead to doubling of the construction speed.

The bill of quantities data may be measured in number, length, area, volume, weight or time and would typically be comprised in data pre- pared by the cost consultant or contractor that provides project specific measured quantities of the items of work, identified by the drawings, 3d models and specifications in the construction documentation, related to one or more construction complexes planned to be constructed during the construction period. Preparing a bill of quantities will normally re- quire that the design is complete and that a specification has been pre- pared. The bill of quantities is typically issued to tenderers in order to allow them to prepare a price for carrying out the works. The bill of quantities is used by the tenderers when calculating construction costs for their tender. The use of bill of quantities ensures that all tendering contractors will be pricing the same quantities, rather than taking off quantities from the drawings and specifications themselves. Hereby, the use of bill of quantities makes it possible to provide a fair and accurate system for tendering.

Production rate data describe the number of goods that can be produced during a given period of time. In a specific construction pro- cess, this is the rate at which workers are expected to complete a cer- tain segment in a location, such as a building of a single floor of a build- ing. The production rates may by way of example be the number of workers required and man-hour requirement for form works, rebar in- stallation and concrete casting. As work is mostly done in teams, the composition of the team, labor and equipment used will mostly deter- mine production rates. For carrying out the same work, different teams may have different production rates.

The calculation module of the computer that is used for the computer implemented calculation is configured to calculate activity durations of a number of activities in a location-based manner and hereby provide a time schedule. The activities may include concrete construction activi- ties such as concrete slab work, foundation work, wall construction work, studwork, electrical installation work, including electrical rough-in and electrical wiring, installation and mounting of doors, windows, in- stallations of plumbing, including system of pipes, drains fittings, valves, valve assemblies, and devices installed in a building for the dis- tribution of water for drinking, heating and washing, and the removal of waterborne wastes, as well as tape and finish work.

For example, the method is performed in a centrally arranged computer unit. In another embodiment, the calculation module is based on cloud computing, at least partially. Cloud computing may be beneficial be- cause it does not require without direct active management by the user.

The time schedule includes all information necessary to carry out the construction process, for example building construction process. The time schedule includes all required activities, including materials and staff to carry out the activities.

The optimization module makes it possible to optimize new time sched- ules due to change in resources, requirements, and conditions, in order to save time and costs. By updating the time schedule due to re-run- ning the calculation, the system is capable of providing a more realistic plan by the multi-iterative calculations and predict the total cost and re- maining construction time better during the construction period. For ex- ample, the optimization module is configured to calculate a time sched- ule in a manner in which a number of not only internal but also external logical relationships between activities within locations are taken into consideration. Optionally, a number of external higher-level logical rela- tionships between activities driven by different levels of location accu- racy are taken into consideration. In some embodiments, the optimiza- tion module is configured to calculate a time schedule by including phased hybrid logic between tasks in related sub-locations.

Advantageously, the optimization module is configured to identify con- struction specific dependencies automatically by using information from a model module about the structures forming the construction struc- tures of the locations.

Advantageously, the optimization module is configured to calculate a time schedule on a real-time basis. Optionally, the optimization module is configured to even recalculate a new time schedule on a real-time ba- sis. Hereby, it is possible to identify and present to one or more uses of the system an optimized time schedule on a continuous and real-time basis. In practice, the optimization module secures that an optimized time schedule is available at all times. The construction scheduling sys- tem according to the invention is capable of recalculating a new time schedule using any received information that may affect the present time schedule.

Mathematically, the optimization module of the construction scheduling system may use various optimization criteria and constraints for calcu- lating an optimum time schedule. In one embodiment, the optimization criterion is based on minimizing the total cost. In an embodiment, the optimization criterion is based on minimizing the duration of the construction period (minimization of completion time). These two crite- ria are typically combined.

In practice, optimization is carried out by finding a local minimum in a N-dimensional space spanned by a plurality of parameters, on which the time schedule depends. The optimization is, however, setup with the purpose of identifying a global minimum of the N-dimensional space.

In one embodiment, the optimization module used in the system or in the method according to the invention is configured to calculate a time schedule and/or recalculate a new time schedule on the basis of a rein- forcement learning approach. Optionally, it uses a multi-layered logic, for example three, four or five layered logic that implements location- based logic.

In contrast to activity-based planning that works with a one-layer logic in the implementation of the critical path method, location-based man- agement advantageously follows a five-layered logic that implements location-based logic. The layers are as follows and are not hierarchical and apply equally:

1. External logical relationships between activities within locations.

2. External higher-level logical relationships between activities driven by different levels of location accuracy.

3. Internal logic between activities within tasks.

4. Phased hybrid logic between tasks in related locations.

5. Standard critical path management links between any tasks and dif- ferent locations.

It may be advantageous that the optimization module is configured to redistribute the resources. Hereby, it is possible to use the resources in the most optimum manner. By way of example, if a first notification of illness of a first carpenter working in a two-man team in a first location and a second notification of illness of a second carpenter working in a two-man team in a second location is received one morning, it may be more efficient that two carpenters that are not sick work together in the same locations in order to work more efficient (some carpenter work re- quires two person). Accordingly, it may be a major advantage that the construction scheduling system is configured to automatically redistrib- ute the resources based on a predefined optimization criterion. In one embodiment, the predefined optimization criterion may be to minimise the total cost. In another embodiment, the predefined optimization cri- terion may be to minimise the duration of the construction period (mini- mization of completion time). When both criteria are used in combina- tion, predetermined balancing criteria are implemented, as the minimi- zation of time may not lead to minimization of costs, why certain bal- ance criteria have to be applied, for example achieving a reduction of 20% of the costs by accepting a 10% increase in construction time, or vice versa.

By way of example, if a delayed consignment (e.g. delivery of brick) is received one morning, it is possible to allocate the bricklayers to an- other location, in which bricks are already available or to transport bricks (that have already been delivered) from one location to another location. Hereby, the brick layers can work efficiently even if the bricks are not delivered in the scheduled location.

Optionally, the optimization module is configured to receive manually entered input so that manually control is enabled. Hereby, the optimiza- tion module can be controlled by the user. This may be beneficial in sit- uations, in which a user input will affect the building process. By way example, a user input may be used to select one of several options that are equally suitable. By example, a user input may be used to prioritis- ing that the construction work of a specific location is terminated. This may be beneficial if the termination of this location is of political im- portance, for example if the termination of this location increased the chances to get a funding. A manual control is a way to enable the user to select how to proceed.

In some embodiments, the optimization module is configured to provide a plurality of alternative time schedule options for the user. Hereby, the user can select from a list of options with different reductions of the to- tal cost on the expense of various longer duration of the remaining con- struction period.

Optionally, the optimization module is configured to provide a new time schedule being optimised with respect to carbon dioxide (CO2) emission.

For example, the optimization module is configured to change the order of locations and calculate the effect, the duration of the construction pe- riod and the total cost. This is relevant if fewer or additional resources than scheduled are available.

Optionally, the user may initiate the change of the order of locations by entering a user input. In one embodiment, the system is configured to receive such user input. In one embodiment, the system is configured to recommend/suggest such changes without any preceding user input.

To avoid continuous schedule to use non-allowable distribution/alloca- tion of resources, one may introduce a penalty.

It may be an advantage to apply a portfolio-controlled approach rather than project-controlled approach. If a construction process is carried out in several projects, cities or even in different countries at the same time, it may be advantageous that several locations can be controlled by using the same system for scheduling a construction process. If con- struction works are available on one location, these construction work- ers may be used elsewhere. The system for scheduling a construction process is configured to provide several options/suggestions that the users of the system can select.

The method according to the invention is a method for construction scheduling, wherein the method comprises the location-based manage- ment of scheduling a construction process during a construction period, in which one or more construction complexes is constructed, each com- prising a plurality of locations, wherein the method comprises the step of: applying a model module comprising information about the struc- tures forming the building structures of the locations; applying a database configured to receive and store information uploaded by a number of providers/users, wherein the information includes one or more of the following: resource data, cost data, productivity factor data, bill of quantities data and production rates data; calculating activity durations of a number of activities in a location- based manner and hereby provide a time schedule, wherein the method comprises the step of calculating a time schedule in a manner in which a number of construction specific dependencies are taken into consideration.

Hereby, realistic construction schedules are provided for managing a construction process, in which the total cost and the duration of the construction period can be predicated at any point of time of the con- struction process.

Description of the Drawings

The invention will become more fully understood from the detailed de- scription given herein below. The accompanying drawings are given by way of illustration only, and thus, they are not limitative of the present invention. In the accompanying drawings:

Fig. 1A shows a graph depicting planned work versus worked car- ried out as function of time;

Fig. IB shows an example of the location-based scheduling;

Fig. 2A shows a setup of a multiagent reinforcement learning scheme displaying how agents interact with the environ- ment alongside fellow agents in determining actions to take based on states visited during iterative system runs;

Fig. 2B shows a table of bi ll-of-quantities data of a sample environment;

Fig. 3A shows an example of location-based dependencies in a given precision level (Project, Building, Level) on the de- pendency between T004 and T005. Only locations of a building A and a building C;

Fig. 3B shows a Multiagent Reinforcement Learning Algorithm ac- cording to the invention;

Fig. 4A shows a location-based management (LBM) Control Algo- rithm according to the invention;

Fig. 4B shows an example of the activity agent design and its ac- tion space;

Fig. 5A shows a final diagram view of a single agent in an activity- on-node (AON) network, wherein the diagram is split be- tween the agent view, and the control algorithm view as the counterpart;

Fig. 5B shows an example of a schedule optimization result;

Fig. 6A shows an illustration of a building complex;

Fig. 6B is a table listing calculated quantities of selected structures of four selected locations of the building complex shown in Fig. 6A, wherein the quantities are calculated on the basis of the three-dimensional model module corresponding to the building complex shown in Fig. 6A;

Fig. 6C is a table illustrating components of the structures listed in Fig. 6B and

Fig. 7 shows an example of how a system according to the inven- tion visualises selected locations and simultaneously present process data related to the selected locations;

FIG. 8 is a flow diagram for computer implemented planning.

Detailed description of the invention

Referring now in detail to the drawings for the purpose of illustrating preferred embodiments of the present invention, a graph 28 depicting planned work 24 (indicated by a solid line) versus worked 26 carried out (indicated by a dotted line) as function of time is illustrated in Fig. 1A.

It can be seen that the work 26 carried out has been completed a time period At before scheduled. Accordingly, the construction process is ahead of schedule. The system and method according to the invention is capable of taking advantage of this situation by redistributing the available resources, especially the workforce, including the available working hours of one or more craftsmen, to another location. Hereby, it is possible to utilize available resources elsewhere and thus accelerate the construction process.

Fig. IB illustrates an example of the location-based scheduling. Intro- ducing duration as the only optimality measure implies the risk of unde- sirable resource allocation, as it is not implicit for location-based sched- uling with paced activity settings that the more resources assigned to each activity the shorter the execution time of the whole schedule. Thus, translating duration into a cost optimization problem while also introducing an efficiency loss to resource/crew management is desira- ble.

Fig. 2A illustrates a setup of a multiagent reinforcement learning scheme displaying how agents interact with the environment alongside fellow agents in determining actions to take based on states visited dur- ing iterative system runs. Being alongside supervised and unsupervised learning as one of three main machine learning archetypes, reinforce- ment learning differs in that the modelling does not ought to minimize prediction errors in labelled datasets. Instead, reinforcement learning focusses on mapping states to actions, while balancing exploration and exploitation of a given environment, and maximizing the numerical re- ward signal. Agents ought to learn optimal problem-solving heuristics of the environment, have no prior knowledge of the environment and are to gain knowledge of the environment by exploring which actions yield the highest accumulated reward signals and storing such learnings in memory. The problem is that neither exploration nor exploitation can be pursued fully without failing the task.

Though simpler reinforcement learning setups only work with one interacting agent, a solution for the resource allocation tool is to intro- duce multiple agents to the same environment mimicking a real-world negotiation among the project stakeholders.

The idea of a multiagent system is simple in that its agents hold the ability to take independent actions in order to satisfy its design objec- tives and to do so in interaction with multiple agents in the same envi- ronment. A multiagent system learning scheme can be seen in Fig. 2A.

Good multiagent system designs offer the ability of agents to cooperate, coordinate and negotiate with each other in finding the optimal joint so- lution to the problem at hand, hence likely applicable to project man- agement and more specifically re-source allocation following theories of Last Planner System (suggested by the Lean Construction Institute) and location-based management.

Agents work from an initial state where full environment details are available to all interacting with the system. Globally exerted variables of the system include a productivity decay factor that integrates crew effi- ciency to the system based on its composition. Variants of the decaying factor include exponential decay, gamma-distributions, chi-square dis- tributions and mixture distributions etc. productivity decays can refer to certain activity and/or resource.

Variables used in the multi agent reinforcement learning algorithm en- compass an early stopping mechanism guarding a maximum iteration count, which triggers a termination of the else infinite optimization search. The variable triggering the search termination is found from empirical data to be in the span of 100-500, for example 200-400, with regards of the schedule complexity, although 100 iterations can be suf- ficient in some cases.

A variable guarding for the learning rate of the algorithm (the rate at which the system adopts to the environment) is found to be crucial in allowing the algorithm to succeed. A learning rate below 0.1 is desirable as going higher will prompt unstable results.

The algorithm moreover encompasses a variable guarding the search randomness. This variable is of minor importance and is set to be of 0.5 probability to giving the control to the single random dispatcher of the system.

Fig. 2B illustrates a table of bi ll-of-quantities data of a sample environ- ment.

In the following, suitable project methods are described in detail. This includes the environment setup, the model design, and evaluation setup. The methods are described while considering the context of loca- tion-based scheduling, reinforcement learning, and the multimode re- source-constrained project scheduling problem as well as the project objectives.

The project prototype can be built using Python 3.6, Jupyter, and the use of open-source libraries. Included libraries are NetworkX for crea- tion and manipulation of complex net-works, Pandas for handling data structures and data analysis, NumPy for list operations, Matplotlib for plotting, tqdm for progress status handling and dill for exporting gener- ated results. Furthermore, the setup in NetworkX may be based on find- ing the critical path with topological sorting. Though, the simple critical path method-setup was extended to encompass theories of location- based management by adding location-based logic layering of the prec- edence setup as well as adding new logical layers to the schedule gen- erating scheme.

A sample environment/schedule was established from a demo construc- tion project in collaboration with an engineering consultancy firm. The schedule outline was made based on location-based bill-of-quantities, productivity, and production rates in establishing a setup for calculating activity durations in a location-based manner. Moreover, relating to the activity durations are the resource allocation. Base resource allocations are initialized at random.

The sample environment consists of 16 locations, including the project itself, in a three-level deep hierarchical location breakdown structure. The location hierarchy levels are ranked 1. Project, 2. Building (A, B, C), 3. Building Level (1, 2, 3, 4, 5). The based bill-of-quantities is held at a minimum complexity accounting for gross floor area with m ² as the unit of measure.

To add to the simplicity of the schedule outline, each activity only holds one quantity as an activity driver, which is then influencing the dura- tion. Table 1 shown in Fig. 2B illustrates an excerpt of the based bill-of- quantities on activity type level, making the basis of the project. The sample projects hold 21 unique activity types specified with a code (e.g., T005) and a name (e.g., Exterior Walls).

Activity durations are calculated based on the bill-of-quantities, produc- tivity, production rates, workday/shift length, and the number of re- sources allocated. The total duration of a location-based activity is given by the following equation:

Where: A vital modification to note is the addition of (P to the power of r ^M ). The production efficiency P refers to the deterioration of production effi- ciency when adding more resources to the activity. The efficiency loss is justified by the need for extra resource management when adding more resources to the task. The efficiency is presumed to follow an exponen- tially decaying function with the base Pε[0,1]. For this sample environ- ment, P=0.98 is deemed reasonable.

The efficiency loss just described is directly linked to the total project cost, which is functioning as the backbone of the reward signal of the model. By adding the efficiency loss to the overall economy representa- tion, inexpedient and undue resource allocations can be avoided.

The location-based durations are one of three primary information's needed to establish the sample environment. Described in the next sec- tion is the two remaining primary information's namely, resource con- straints and activity dependencies.

Location-based dependencies are best explained by an example. Take task T004 (Exterior Walls) and T005 (Slabs) in Fig. 3A. In standard CPM settings if T005 is a successor of T004, works of T005 in the project are only allowed to begin after completion of T004.

The solid black lines make up the necessary location-based dependen- cies in a given precision level (Project, Building, Level) on the depend- ency between T004 and T005. Only locations of Building A and C are il- lustrated in this example explaining the system.

In a location-based setting, successor tasks are allowed to start after completing the location given the precision assigned to the dependency. These settings refer to the external activity logic, while internal location logic applies as well as determining the order of locations for each activ- ity type. The internal logic secures the right internal flow and can be set to be paced or non-paced (discontinued). For simplicity reasons, the in- ternal location order is kept the same for all activity types. This, how- ever, limits the optimization space. Whether an activity type is allowed to be discontinued is set by the scheduler. The above-shown figure demonstrates how multilevel dependency structures link together.

Each activity type can be executed in a set of KaR resource modes. Each mode represents a different combination of requested re- sources to the activity type; how-ever, the current model only considers one renewable resource type accusation per activity type from a set of R renewable resources representing the different tradesmen of the con- struction project. The renewable resources rε{1,..., R} have constant availability for each time-period throughout the project.

A solution to the scheduling problem must contain location-based start and finish times of the activities and must present such times while complying with overall constraints. The objective is to find a resource mode for each activity type minimizing for the overall cost of the project C ^T , presented in equation (2) shown below, while not exceeding the re- source availability. The optimization criteria also reckon in the total pro- ject execution time. The project execution time is not suitable in itself for location-based management, as it can lead to inexpedient and undue resource allocations. The system will contain designs of activity agents and a single dispatch agent. The former will be designed as an agent per activity type. The latter is added to handle agent control as a random dispatcher.

Activity type agents will be designed to take autonomous actions for so- cial welfare optimization. All activity type agents working in the multia- gent system are equipped with a component for mode selection and learning. In the process of learning appropriate schedule heuristics, a simple learning automate technique is applied. The goal is to find an op- timal action based on an environmental response from previous actions made by the agent, in other words, explained as learning in a trial and error approach.

To store the environment feedback, each agent holds a probability vec- tor p with probabilities of all actions/modes in actions set A. At each in- stant k, k =1,2...,K when running the learning algorithm, the agents choose an action from A with probability exemplified in the following equation :

( ³ )

When an environment response is given, on the action performed by the agent in control, a binary reward re{0,l} is given. The reward sys- tem is described in detail in section 3.2.2. The reward is fed to a learn- ing algorithm scheme U in the following equations (4) and (5) responsi- ble for updating the probability vector of each agent.

(4) if α is the action taken at time k else update by

(5) Where:

The adopted scheme for this project involves setting α _p equal to 0 mak- ing the scheme a linear reward-inaction learning algorithm. The L _R-I is proved to be absolute expedient and thus ε-optimal.

Adding to the responsibilities of the activity type agents, they are fur- thermore responsible for keeping information flowing to location-based activities in the nested graph system. Changes in resource allocation on activity type level are to be communicated further downstream to do calculations on new location-based schedule durations. Hence a method is added to the agents to update its location-based children.

For handling agent control, a dispatch agent is introduced and described in the following section. However, to keep a sense of schedule continu- ity in the control flow, a probability of the current agent in control to hand over control to a random successor activity type is added by p _ToDisp . If a randomly generated number between 0 and 1 is higher than p _ToDisp and one or more successors are found, then control is given to the ran- domly chosen successor agent. Else, control is given to the dispatcher agent.

The single dispatcher agent is in charge of global agent control admin- istration and is designed to randomly assign control to eligible agents when control is handed over. The dispatcher ensures a random explora- tion of the activity types in the schedule.

The learning approach is as previously mentioned agent-based and is set to be model-free with a stochastic on-policy paradigm. The agents interact in an environment with a static topology, discrete action space, and continuous state space. The purpose of the interaction design is that the environment is only sensitive to the action performed by the agent in current control and that the agents in the environment work from a social interest incentive. The incentive of social interest results in the maximization of social welfare, which is tied to the overall project economy.

The learning algorithm of the multiagent system is outlined in Algorithm 1 illustrated in Fig. 3B and described in detail in the following. The algo- rithm starts by initializing iterate-counts i _im and i _es and giving control to a random eligible activity type by calling the dispatcher agent. Subse- quently, a while loop is initiated bounded by a maximum iteration count and early stop-ping mechanism that must set in whenever the algo- rithm converges, hence no changes are made to the allocation of re- sources. The limit for when convergence is reached is the preset varia- ble earlyStopping.

The graph that holds the location-based activities is updated on the first iteration. The update is done to carry out a full calculation and update on overall project economy and execution time. After running the up- date, the agent currently in control does a mode selection from its prob- ability distribution, which is starting uniformly. Hence no preconstructed knowledge is fed into the system. Given the action taken, the probabil- ity distribution gets updated on the reward signal sent as feedback from the environment.

The reward signal is made up of a binary learning response that sends a response of 1 if the total project economy is reduced and if the resource constraints are not violated by the action performed by the agent in current control (favorable solution). Else, the reward given as feedback is 0 (unfavorable solution). The reward is based on the location-based control algorithms in place for location-based planning control and re- source constraint control. The control algorithms are described in fur- ther detail in the following. The measure accounting for the rewards is highly susceptible to the ac- tions performed by the agent in control. On this account, the reward signal given if no change in the measure is happening from the action performed is set to be an unfavorable result, hence a 0-reward.

After updating the probability distributions of the action sets, the action is either kept as the new mode for the current activity type or rejected if the resource constraints are violated. If the resource constraints are violated, then mode settings are kept with no changes to them.

Lastly, the control is passed on as previously mentioned, and the itera- tion counters are updated according to the reward signal.

For each feedback loop in learning optimal schedule heuristics, a loca- tion-based management control algorithm is run to construct a schedule from actions taken, also referred to as resource allocations. The loca- tion-based management control algorithm is a schedule generation scheme calculating the total execution time of the project allowing for further usage in calculating the total project economy.

The LBM control algorithm is based on both the directed activity type graph and the nested location-based graph. As previously covered, the activity types form the mainGraph iterator in which the agents operate and also works as the starting point of the location-based management control algorithm outlined in Algorithm 2 shown in Fig. 4A and described in further detail in the following.

Based on the mainGraph consisting of activity types presented in a di- rected activity-on-node (AON) graph, a layered feedforward for-loop is constructed over a topologically sorted iterator. For each activity in the mainGraph a subgraph is defined from location-based activities according to the location specifics of the activity. Furthermore, a nested for-loop is constructed over the location-based activities accounting for external node relations. For each IbActivity in the defined subgraph the earliest start (es) and earliest finish (ef) time are calculated and stored on the graph node.

To take in to account the continuity of each activity an internal logic is imposed to the system accounting for the location order. If the activity is set to be paced, a for-loop over a reverse topological sorted supgraph is initiated. This rule implies internal location continuity by right-justifi- cation in a backward approach.

However, the simple critical path management-setup was quickly ex- tended to encompass theories of location-based management by adding location-based logic layering of the precedence setup as well as adding new logical layers to the schedule generating scheme.

The resource control algorithm covers the regulation of resource usage. Allocated re-sources are checked for violation of settled resource con- straints. The control algorithm functions as a strict binary violation checker where a violation outputs 0, why a feasible solution must out- put 1. All resource constraints that are set in the different trades must be obeyed to reach a feasible solution.

In constructing the resource control signal es and ef times are gathered in a vector outlining resource time bins. By list operations, resource us- age counts are aggregated on resource type and time bins by sum. The binned resource type counts are then checked for violation of global re- source constraints.

For evaluation of the model, the prototype process and development re- sults will be reported in a mixed qualitative and quantitative manner. The solutions of the model-build will be described along with computa- tional gains. Following the results of the prototyping process, the sys- tem enters a finetuning process of hyperparameter tuning regarding αr (reward parameter) and pToDisp (probability of dispatcher control). The hyperparameter search will be conducted in a two-dimensional grid search with arε{0.01, 0.03, 0.05, 0.10, 0.20, 0.40} and with pToDispε {0.25, 0.50, 0.75}. These two parameters are of special interest, where the prior holds the ability to set the rate of learning, which is thought of as crucial in a system without explicit exploration phase. The latter holds the capability of learning the importance of co- herence and schedule continuity in the iterative search. Furthermore, pToDisp also controls the uniformity of the schedule exploration, as lower probabilities of giving control to the dispatcher will favorize down- stream activities in the network. Henceforth, the early stopping scheme is designed to be evaluated by a search amongst different convergence tail lengths. In practice, this is implemented by adjusting the stopping criteria of continuous zero reinforcements.

The stopping criteria earlyStoppingε {100,150,200,300,500} will be searched to give an understanding of how the convergence mechanism works for the constructed model at hand. The results are averaged over three searches. Moreover, a backstopping mechanism was implemented to guard for non-converging and slow-converging learning runs.

No known studies can be found to be comparable to the study at hand. Therefore, there is no known optimal solution of the sample environ- ment presented. Instead, the search of optimality strives to reach an acceptable solution. The search is therefore carried out in what can be described as an undisclosed environment where the search of optimality is satisfied through a search of suboptimality and an acceptable result compared to the best-known solution.

The efficiency of the prototype was evaluated through reiterated model runs with empirically obtained parameters from the hyperparameter search and the convergence analysis. The model will undergo a number of iterations (e.g. 5x1000), 5x5000 iterations and additionally 5xn itera- tions where n is the iterations needed for convergence. As the optimal solution to the problem is not known, the average percentual deviation from the best-known solutions created from domain knowledge is con- sidered.

Fig. 4B illustrates an example of the activity agent design and its action space. There is a need for negotiation-based approaches in Last Planner System and location-based management to reach agreed schedule re- sults amongst supervisors and project managers. Results may be based on stakeholder compromises when negotiating with project partners in reaching the common goal of an optimal schedule.

Fig. 5A illustrates a final diagram view of a single agent in an activity- on-node (AON) network, wherein the diagram is split between the agent view, and the control algorithm view as the counterpart. Work at vari- ous locations is depicted as function of time.

When control is handed over to an agent, then the agent chooses an ac- tion from its set of actions with a probability given in the actions proba- bility vector stored in agent memory. The agent makes an inquiry with the control algorithm to prove whether or not the action is allowed and if it improved the schedule with regards of the optimization criterion('s). The feedback sent from the control algorithm is then used by the agent to update its action probability vector, before handing over the control to either the next agent or the dispatcher.

Fig. 5B illustrates an example of a schedule optimization result. The up- per part shows the "unoptimized schedule". It can be seen that the total duration of the construction period is rather long.

The middle part shows the "optimized schedule". It can be seen that the total duration of the construction period is shorter than in the unopti- mized schedule case shown above.

The lowermost part shows the "best known schedule" found by using the method or a system according to the invention. It can be seen that the total duration of the construction period is even shorter than in the optimized schedule case shown above.

Moreover, many of the lines (each representing work or activities car- ried out) are shorter. Accordingly, the lines have slopes that are steeper. This means that more work/activities have been carried out during a shorter period of time.

Fig. 6A illustrates a building complex 20 comprising several locations L1 L 2, L3, L 4.

Fig. 6B is a table illustrating the content of selected structures of four locations L1, L2, L3, L of the building complex shown in Fig. 6A. In the first column (from the left) a number of structures are listed. These structures are: doors, wall 1 (a first wall type), wall 2 (a second wall type), floor and cable trays.

In the second column (from the left) the total number of structures are listed. In the next four columns the number of structures for each of the four locations L ₁ , L ₂ , L ₃ , L ₄ of the building complex are listed.

Fig. 6C is a table listing components of the structures listed in Fig. 6B. Comparing Fig. 6B and Fig. 6C, it can be seen that the wall 1 is linked to a concrete structure. Likewise, the wall 2 is linked to the concrete structure. The floor is also linked to the concrete structure. The cable trays are linked to the cable tray type 1 structure. The walls (both type 1 and 2) are linked to the interior painting structure.

The floor structure in Fig. 6B is linked to the interior structure.

Fig. 6C shows the quantity of structures for the four locations L ₁ , L ₂ , L ₃ , L ₄ of the building complex. Fig. 6C also shows the unit cost (in the third column from the right), the total cost (in the second column from the right) and the man-hours per unit (in the first column from the right). Accordingly, the information in Fig. 6C can be used to calculate the cost as tasks are being carried out. It can also be calculated whether more staff is needed.

Fig. 7 illustrates an example of how a construction scheduling system 2 according to the invention visualises a building complex 20 comprising a number of locations L ₁ , L ₂ , L ₃ , L ₄ , L ₅ , L ₆ and simultaneously presents process data 14, 14', 14" and a graphical representation 19 for the lo- cations L ₁ , L ₂ , L ₃ , L ₄ , L ₅ . It can be seen that the visualisation 19 is pro- vided by using a screen, in which the locations are shown in an ex- ploded view. Hereby, the user can see details that cannot be seen by using prior art systems. Such details may e.g. be a basement that nor- mally would be impossible to see because a floor would be provided on top of the basement.

The construction scheduling system 2 has a plurality of providers 10, 10', 10' and users 12, 12'.

The construction scheduling system 2 comprises a three-dimensional model module 6 comprising information 8 about the structures forming the building structures of the locations L ₁ , L ₂ , L ₃ , L ₄ , L ₅ , L ₆ .

The construction scheduling system 2 comprises a database 16 config- ured to receive and store information uploaded by a number of provid- ers 10, 10', 10' and users 12, 12'. The information 8 may be resource data, cost data, productivity factor data, bi ll-of-quantities data and pro- duction rates data.

The construction scheduling system 2 comprises a calculation module 22 configured to calculate activity durations of a number of activities in a location-based manner and hereby provide a time schedule 20.

The construction scheduling system 2 comprises an optimization module 18 configured to recalculate a new time schedule 20 when one or more of the information 8 is changed, wherein activity durations are calculated in a location-based manner. In Fig. 7 the optimization mod- ule 18 is integrated in the calculation module 22. In another embodi- ment, however, the optimization module 18 may be provided as a sin- gle structure separated from the calculation module 22.

In one embodiment, a number of providers 10, 10', 10" providing any suitable service that may include any of the following: design, engineer- ing, bricklaying, carpentry, piping, mechanical installation, painting, fur- nishing and work carried out by other professions such as electrical technicians, can access the database 6 in order to enter data or update data. Such data may include, estimated working hours, actually spent working hours, predefined milestones for performing tasks or sub-tasks thereof (e.g. laying bricks or installing a pipe system), material used, material to be used, material cost (e.g. cost per area or volume). The users 10, 10' shown in Fig. 7 may be managers seeking information e.g. about the status or remaining cost of the construction process.

The process data 14, 14', 14" may be one or more of the following:

- the estimated remaining construction time and/or;

- the cost up to the point in time and/or;

- the work accomplished measured relative to the scheduled work and/or;

- resources such as manpower, equipment and machinery, and con- struction material.

FIG. 8 illustrates an overview of the computer implemented iterative process 100.

The step 101 represents the provision of updated construction progress data, representing the actual construction state. These data serve as in- put for the offset of the iterative process 100. At the first iterative cal- culation, the construction state 101 is at a start phase, but as the con- struction progresses, the construction state 101 changes, which is im- plemented by change is progress data 101, which is subsequently used for new, updated calculations during the construction. A counter in step 108 keeps trace of the number of iterations.

The step 102 represents the provision of a time schedule for the con- struction. This time schedule is dependent on the already achieved pro- gress and may be revised during the construction on the basis of the in- put from the update construction progress data from step 101.

The step 103 represents a selection of a task (Agent) for the next round of optimization. The selection is done either by the dispatcher or chosen among the successors of the previous task . The method for choosing a successor, if such exists, would be to choose at random using a weighted probability taking into account task duration, task cost and re- source efficiency, such that the more important tasks, by these metrics, will be chosen more often. The dispatcher would choose a task com- pletely at random to start from again, and if no successor exists the dis- patcher will be activated in this manner as well. If some of the con- struction activities at sub-locations have been performed faster than originally calculated, see also FIG. 1 in this respect, the order and start time of subsequent tasks are potentially adjusted for optimization. For example, some tasks at certain sub-location may be able to start earlier than originally planned. The selection of tasks is prioritized on the basis of a priority input from step 109, which will be explained in greater de- tail below.

The step 104 represents a decision routine on the task selection of 103. This step 104 implies workforce selection, workforce allocation, as well as resource requirements. This choice will consider a probability distri- bution that changes during the optimization, more precisely step 109, to reflect the effect of past choices.

The step 105 represents adjustment of the parameter in question to the value selected in step 104, for the task selected in step 103. The step 106 represents a further adjustment of the construction schedule by implementing a multi-layer logic, as explained above, in or- der to make sure that the adjusted plan is still within the requirements and constraints for the model. If the constraints cannot be followed, the plan may not be an improvement compared to the previous plan. This is evaluated in the next step 107.

The step 107 evaluates the effect of the adjusted construction schedule of step 106. For the evaluation in 107, resource data are received in step 111 as input, as the amended plan has to be checked in relation to the available resources relatively to the required resources as per the amended plan. Potentially, the step 111 delivers input data of available resource data as well as resource data that could be made available, if it has high priority.

If the effect is advantageous, in particular by reducing costs and/or time, a positive evaluation parameter is forwarded to an artificial intelli- gence routine 109 as a reward, as this indicates a positive direction of optimization in the iteration process. If the effect is not advantageous, for example because the construction gets more expensive and/or takes longer time, or if the requirements for resources in the new plan cannot be covered by existing resources as per availability in 111, a negative evaluation parameter is forwarded to an artificial intelligence routine 109, indicating that this direction of iteration is not useful, prompting the Al routine 109 to use another direction in the iteration process. The Al routine maps the evaluation parameters for positive and negative ef- fects and uses it as input to the task selection routine in step 103.- The latter implies that the Al routine learns about good and bad directions in the iteration process.

The routines circuit is repeated a predetermined number of times, and a counter-step 108 keeps track on the number of iterations. The number of iterations has to be selected such that there are sufficient iterations for the Al routine 109 to find a convergence direction towards a beneficial optimization of cost and/or time. The first few iterations may not lead to any useful optimization and not even to a specific direction for optimization, but after a certain number of iterations and by evalu- ating in step 109 the evaluation parameter from step 107, the overall computer implemented iterative method will come closer to a conver- gence towards a local or global minimum of cost and/or time for the construction process.

As long as the predetermined number of iterations has not been reached, the counter step 108 continues to order further iterations through the module 109. When receiving such order of further itera- tions, the Al trained model in step 109 continues the optimization on the basis of the received and mapped evaluation parameters, each of which is related to different modelled plans during the iterations.

When the predetermined number of iterations has been reached, the counter step 108 stops the process and triggers an output 112 of the new plan. The fact that the predetermined number of iterations has been chosen large enough, for example in the range of 100-500 itera- tions, a convergence towards an optimized change of the construction schedule can be assumed so that the output can be expected not to yield unfortunate or even useless results.

Periodically, for example weekly or daily, such an iterative calculation process is repeated. For such repeated process, the input progress state in step 101 is adjusted as an offset, and the resource data are poten- tially adjusted in step 111, in order to check whether necessary re- sources are available and can be used for the optimization. Due to the mapping of evaluation parameters and the Al learning process in 109, earlier disadvantageous adjustments, which led to a negative evaluation parameter from step 107, are also taken into account in step 109 so that the Al learning continues throughout the entire construction pro- cess, with increased efficiency and reliability as the construction pro- gresses. In alternative embodiments to the iteration being performed until a maximum predetermined number of iteration has been counted in 108, the iteration process is stopped when a satisfying construction schedule has been found due to cost and/or time requirements and when the changes along a certain direction of the parameters do not lead to any further substantial improvement. This way, the calculation time can be shortened. In this case, the stopping due to a satisfying plan is overrul- ing the counting criterion.

List of reference numerals

2 Construction scheduling system

4 Construction complex (e.g. building complex

6 Model module

8 Information

10, 10', 10' Provider

12, 12' User

14, 14', 14" Process data

16 Database

18 Optimization module

19 Visualisation

20 Time schedule

22 Calculation module

24 Planned work

26 Work actually carried out

28 Graph

L ₁ , L ₂ , L ₃ Locations

L ₄ , L ₅ , L ₆ Locations

100 computer implemented iterative process

101 provision of updated construction progress data

102 provision of a time schedule for the construction

103 selection of tasks for the next group of activities

104 decision routine on potential activities

105 adjustment of the existing construction schedule

106 further adjustment of the construction schedule by im- plementing a multi-layer logic

107 evaluation of effect of the adjusted construction schedule

108 counter step

109 artificial intelligence routine

110 update of plan

111 input of available resource data

112 output