FAULT DETECTION IN HVAC- SYSTEMS USING BUILDING INFORMATION

MODELS AND HEAT FLOW MODELS

CROSS REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S.

Provisional Application No. 61/252,862, filed on October 19, 2009, the disclosure which is incorporated herein by reference in its entirety.

BACKGROUND OF THE INVENTION

[0002] The invention relates generally to process control. More specifically, the invention relates to a Heat Flow Model (HFM) methodology for HVAC system fault detection and diagnosis. Embodiments use the modularity of graphs to achieve a direct automated mapping of HVAC structures and components into HFM graphs, and use node behavior model and software libraries to translate the HFM graphs into systems that can be integrated in HVAC control systems.

[0003] Modern Heating, Ventilation and Air Conditioning (HVAC) control systems are often too complex to have proper means for effective Fault Detection and Diagnosis (FDD) and correction. With many existing FDD approaches, the engineering efforts to apply and adapt them to various HVAC systems are great.

[0004] Building HVAC mechanical and control systems are prone to many faults that cause failures. Some failures lead to alarms, others decrease energy efficiency, lifetime of the system and user comfort without obvious notifications. Even if failures are detected by a maintenance staff, fault localization is often very difficult because there is no one-to-one correspondence between faults and reported failures. [0005] FDD for HVAC has become an important topic with many contributions from academics and industry. However, one problem still remains when the research results are put into practice, since nearly all buildings are different there is a huge variety of HVAC systems. If the FDD system does not match a particular HVAC system, not enough faults are detected. And if it is customized for a specific system, the development effort and cost may be too high in relation to the possible gain.

[0006] The use of expert systems to diagnose faults in HVAC components and systems have been tried and include rule-based methods, fuzzy model based strategies, and Artificial Neural Network (ANN) based classifiers.

[0007] Usually a set of failure rules based on

temperature or pressure inequalities is derived to detect faults. In most of the studies, either the rules have been derived manually for each specific HVAC system or the ANN has to be trained offline which may not

necessarily cover all of the faults due to the limited training data.

[0008] These processes are time consuming and labor intensive. As the recent advancement in building modeling technology has already impacted the building design and construction engineering process significantly, it is believed that developing an FDD system based on building information models will both enhance building fault diagnosis capability and reduce the engineering process of generating fault rules.

[0009] Multilevel Flow Models (MFM) have been applied to power plants and similar systems. The flow models are graphs representing mass, energy and information flows. Once the graphs are established, rules based on mass ar energy conservation laws are extracted and analyzed by inference engine to realize the FDD in real-time .

[0010] The MFM flow models are graphs representing mass, energy and information flows where mass and energy conservation laws apply and can be used for fault detection and diagnosis. However, to use MFM, the flows have to be measured, which are often not available for an HVAC system.

[0011] What is desired is a system and method that provides FDD to reduce this effort.

SUMMARY OF THE INVENTION

[0012] The inventors have discovered that it would be desirable to have systems and methods that provide a Heat Flow Model (HFM) methodology. Embodiments automatically translate formal HVAC system descriptions from a Building Information Model (BIM) into HFM graphs, and compile the graphs into executable FDD systems. During an engineering phase, a Graphic User Interface (GUI) is used to

configure parameters, conditions, and logic switches not found in the BIM. During a runtime phase, real-time data from an HVAC control system is input to the generated FDD system (HFM graph) for fault detection and diagnosis.

[0013] Embodiments create a hierarchical HFM graph model which can be used for automatic HVAC FDD generation from appropriate BIMs such as Industrial Foundation Class (IFC). HFM has a one-to-one correspondence with the component structure of existing or planned HVAC systems. HFM nodes such as coils, ducts or fans model the dynamic physical behavior (temperature, flow, humidity and pressure) of air and water flows of these components as precise as can be derived from the BIMs . Each node can calculate its upstream and downstream physical behavior parameter values of connected HVAC components with the dynamic output data from the HVAC control system using instrument sensor and control values. The results are propagated through the HFM graph as parameter ranges.

[0014] Embodiments apply failure rules in each HFM node. If calculated estimate and received value ranges do not match, a fault is presumed and its severity is propagated up in the component hierarchy for diagnosis. The

diagnosis is performed by a central engine by mapping the rule violations from HFM nodes to the failures of the HVAC system, the mapping relationship is represented by an Associative Network. The HFM based FDD is composed of an engineering tool and runtime system.

[0015] Embodiments create an HFM graph based model for building HVAC system FDD. Embodiments automatically extract the necessary structural and quantitative data about the target system from BIM descriptions, e.g. IFC. Rules for detecting faults that are related to nodes of the graph are defined based on first principles.

[0016] One aspect of the invention provides a Heat Flow

Model (HFM) node used in a Heating, Ventilation and Air

Conditioning (HVAC) fault detection graph. Aspects according to the HFM node include a Fwdln edge configured to receive parameter ranges from a downstream direction, a FwdOut edge configured to output parameter ranges in the downstream direction, a Revln edge configured to receive parameter ranges from an upstream direction, a

RevOut edge configured to output parameter ranges in the upstream direction, and node specific configuration data that defines the functionality of the node. [0017] Another aspect of the HFM node is one or more

RulesOut edges configured to output a failure rule decision where the node specific data further comprises a failure rule corresponding to each RulesOut edge where, for the downstream direction, a failure rule compares an estimated parameter range with a Fwdin edge parameter range and for the upstream direction, a failure rule compares an estimated parameter range with a Revin edge parameter range, and if the estimated parameter range is not within the received parameter ranges, a failure is output .

[0018] Another aspect of the HFM node is the node specific configuration data further comprises Fwdin edge parameter range tolerances and Revin edge parameter range tolerances .

[0019] Another aspect of the HFM . node is for the downstream direction, an estimated parameter range is a product of a Revin edge parameter range and a Revin edge parameter range tolerance, and for the upstream

direction, an estimated parameter range is a product of a Fwdin edge parameter range and a Fwdin edge parameter range tolerance .

[0020] Another aspect of the HFM node is one or more

Dataln edges, each configured to receive a dynamic HVAC control system variable, and the node specific

configuration data further comprises Dataln edge dynamic HVAC control system variable parameter tolerance values.

[0021] Another aspect of the HFM node is for the downstream and upstream directions, an estimated

parameter range is a product of a Dataln edge dynamic HVAC control system variable and its dynamic HVAC control system variable parameter tolerance value.

[0022] Another aspect of the invention is a method of using a Heat Flow Model (HFM) graph for detecting HVAC system faults for a building. Aspects according to the method include translating formal HVAC system

descriptions from a Building Information Model (BIM) for the building as HFM nodes, retrieving HVAC component attributes from the BIM for each HFM node, retrieving predefined HFM nodes from an HFM node library, creating connectivity among different HFM nodes from BIM

connectivity data, compiling the HFM nodes into an HFM graph, inputting real-time data from the building HVAC control system to the HFM graph for fault detection, detecting building HVAC system faults using rules defined based on first principles that are related to the HFM nodes, and mapping rule violations from the HFM nodes to the building HVAC control system failures.

[0023] Another aspect of the method is where each node in the HFM graph estimates upstream and downstream physical behavior values that correspond to their building HVAC components with dynamic output instrument sensor and control data from the building HVAC control system and propagates the upstream and downstream physical behavior values through the HFM graph as parameter ranges.

[0024] Another aspect of the invention is a building

Heating, Ventilation and Air Conditioning (HVAC) fault detection system. Aspects according to the method include an interface configured to access a Building Information

Model (BIM) file library and import the building HVAC system BIM files, an HFM node library configured to store a plurality of different predefined HFM node types where an HFM node models the dynamic physical behavior

parameters, of air and water flows in predefined HVAC components as derived from the BIM files, a Graphic User Interface (GUI) configured to input and edit HFM node and linkage configuration data during an HFM graph assembly, a compiler coupled to the interface and GUI, configured to compose together the BIM file data with the additional configuration data, and a Fault Detection and Diagnosis (FDD) generator coupled to the compiler and HFM node library, configured to compare the BIM file types for the building HVAC system with the predefined HFM node types and select HFM nodes that correspond and generate an HFM graph where the HFM graph is by mass air flow path corresponding to the building HVAC system components and behavior .

[0025] Another aspect of the system is an FDD engine configured to instantiate the HFM graph as a runtime system for the building HVAC control system, and an interface configured to access the building HVAC control system, where the FDD engine executes the HFM graph with HVAC control system process and control variable data and applies rules for detecting HVAC system faults.

[0026] The details of one or more embodiments of the invention are set forth in the accompanying drawings and the description below. Other features, objects, and advantages of the invention will be apparent from the description and drawings, and from the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

[0027] FIG. 1 is an exemplary HVAC Fault Detection and Diagnosis (FDD) integration. [0028 ] FIG. 2 is an exemplary Air Handling Unit (AHU) duct and instrumentation diagram.

[ 0029] FIG. 3 is an exemplary Heat Flow Model (HFM) graph .

[ 0030 ] FIG. 4 is an exemplary HFM graph for the AHU shown in FIG. 2.

[0031 ] FIG. 5 is an exemplary downstream temperature range propagation for the AHU HFM graph shown in FIG. 4.

[0032 ] FIG. 6 is an exemplary failure rule situation showing five different calculated estimate tolerance ranges .

[0033] FIG. 7A is an exemplary HFM sensor duct node.

[ 0034 ] FIG. 7B is an exemplary HFM temperature sensor duct node.

[0035 J FIG. 7C is an exemplary HFM flow controlled fan node .

[ 0036] FIG. 7D is an exemplary HFM pressure controlled fan node.

[ 0037] FIG. 7E is an exemplary HFM coil node.

[ 0038 ] FIG. 7F is an exemplary HFM thermostat node.

[ 0039] FIG. 7G is an exemplary HFM mixing box node.

[ 0040 ] FIG. 7H is an exemplary HFM two-way branch node. [0041] FIG. 71 is an exemplary HFM reheat Variable Air Volume (VAV) node.

[0042] FIG. 8 is" an exemplary FDD system using HFM nodal analysis .

[0043] FIG. 9 is an exemplary table showing Industrial Foundation Class (IFC) types corresponding with HFM nodes .

DETAILED DESCRIPTION

[0044] Embodiments of the invention will be described with reference to the accompanying drawing figures wherein like numbers represent like elements throughout. Before embodiments of the invention are explained in detail, it is to be understood that the invention is not limited in its application to the details of the examples set forth in the following description or illustrated in the figures. The invention is capable of other

embodiments and of being practiced or carried out in a variety of applications and in various ways. Also, it is to be understood that the phraseology and terminology used herein is for the purpose of description and should not be regarded as limiting. The use of "including," "comprising, " or "having, " and variations thereof herein is meant to encompass the items listed thereafter and equivalents thereof as well as additional items.

[0045] The terms "connected" and "coupled" are used broadly and encompass both direct and indirect

connecting, and coupling. Further, "connected" and

"coupled" are not restricted to physical or mechanical connections or couplings. [0046] It should be noted that the . invention is not limited to any particular software language described or that is implied in the figures. One of ordinary skill in the art will understand that a variety of software languages may be used for implementation of the

invention. It should also be understood that some of the components and items are illustrated and described as if they were hardware elements, as is common practice within the art. However, one of ordinary skill in the art, and based on a reading of this detailed description, would understand that, in at least one embodiment, components in the method and system may be implemented in software or hardware.

[0047] Embodiments of the invention provide methods, system frameworks, and a computer-usable medium storing computer-readable instructions that translate formal HVAC system descriptions f om a. BIM into a Heat Flow Model

(HFM) graph for HVAC FDD having a one-to-one

correspondence with the component structure of a building HVAC system, which are compiled into an executable FDD system. Real-time data from the building HVAC control system is input to the ' FDD system for fault detection and diagnosis. The invention may be deployed as software as an application program tangibly embodied on a program storage device. The application code for execution can reside on a plurality of different types of computer readable media known to those skilled in the art.

[0048] An HFM is a graph comprised of HFM nodes that correspond to real HVAC system components. The HFM nodes correspond to mass flow connections such as ducts or pipes or electrical energy. The HFM nodes simulate real

HVAC components' dynamic functionality (behavior) as precise as the provided BIM parameters and the dynamic control system data allow for. An HFM graph can be generalized as a component hierarchy. This allows for different levels of refinement and abstraction.

[0049] A BIM is a digital representation of physical and functional characteristics of a facility and serves as a shared knowledge resource for information about a facility. One BIM is the Industry Foundation Classes (IFC) . The IFC modeling language is powerful and can be extended to describe every detail of an HVAC system.

However, there is no unique way to model a specific system. Embodiments use IFC models of existing or planned buildings to create FDD systems automatically.

[0050] IFC is one of the most commonly used formats for interoperability. The IFC model provides a standard representation of the underlying objects, their

properties and relationships including HVAC, electrical, . plumbing, fire protection, building control, etc.

[0051] An HFM based FDD system can be integrated into an existing HVAC supervisory control system structure of a building. The FDD system can communicate directly with top level control. However, an HFM graph representation permits a large degree of modularity of the FDD system and integration at any level of the system as a

distributed system.

[0052] FIG. 1 shows an exemplary HVAC control system 101. The system 101 comprises one zone with one single duct Air Handling Unit (AHU) 103 to supply hot or cold air for two reheat Variable Air Volume (VAV) systems 105ι, 105 _∑ (105 collectively) . Each VAV 105 distributes hot or cold air to a space via outlets with thermostatic feedback control. Return air is collected and returned to the AHU 103.

[0053] FIG. 2 shows the structure and control of the AHU 103. The AHU 103 comprises a return air fan Rfan, a supply air fan Sfan, a mixing box Mixer, a heating coil Hcoil with modulating control valve Hcv, a cooling coil Ccoil with modulating control valve Ccv, four temperature sensors Tra, Toa, Tma, Tsa and one pressure sensor Psa . The supply fan Sfan may be pressure controlled. The return fan Rfan is controlled by the return air pressure to keep air flows balanced. The pressure controlled fan may include a differential pressure sensor to measure the air flow, which is monitored by an AHU controller (not shown) . A VAV 105 typically comprises a damper to control the air flow into a space, a heating coil, a distribution temperature sensor, and an air flow rate sensor (not shown) .

[0054] An HFM comprises two types of data: 1) mass that comprises air and water, and 2) electric energy, and is represented by a graph with nodes having parallel directed and anti-parallel directed edges. The HFM nodes model HVAC components and each node's edges input or output state variable and information flows.

[0055] FIG. 3 shows the parallel and non-parallel edge interconnections of HFM nodes 1 and 2. Parallel Fwdln edge inputs physical parameter data to the node from a downstream direction and FwdOut edge outputs physical parameter data that may be the same or different in the downstream direction of mass flow. Revln edge inputs physical parameter data to the node from an upstream direction and RevOut edge outputs physical parameter data that may be the same or different in the upstream direction. For example, a heating coil node can have separate Fwdin/FwdOut edges and Revin/RevOut edges for air and heating water physical data interconnections if the air and water are modeled.

[0056] The number of anti-parallel Dataln edges and RulesOut edges vary depending on the number of mass or energy flows that are considered in a node. Dataln edges each input a dynamic HVAC control system sensor (process variable) and control (control variable) data. RulesOut edges each output an HFM node specific failure rule value to an FDD engine.

[0057] For each HFM node, the real physical data is represented as a vector. For FDD, air flow data

parameters of interest are temperature, mass flow rate, humidity and pressure.

[0058] In order to generalize HFM node interconnections, a vector with four air flow data parameter ranges is propagated between HFM nodes in downstream A, and upstream B _j directions,

Tmax

Tmin

Qmax

Qmin

[0060] B. 12)

Hmax

Hmin

pmax

pmin

[0061] Each downstream A _j and upstream B _f vector comprise parameter ranges 1) Tmin, Tmax which are the minimum and maximum dry bulb air temperatures, 2) Qmin, Qmax which

are the minimum and maximum mass air flow rates 3) Hmin,

Hmax which are the minimum and maximum water vapor

pressures, and 4) pmin, pmax which are the minimum and

maximum air pressures.

[0062] Heating or cooling water vector state variable

data comprises temperature, flow and pressure values.

[0063] FIG. 4 shows an HFM node graph 401 for AHU 103

comprising eight HFM nodes that include five HFM node

types: two fan nodes (Rfan, Sfan) , two sensor duct nodes

(Rduct, Mduct) , one mixer node [Mixer) , two coil nodes

(Hcoil, Ccoil) and one two sensor duct node (Sduct) . The

HFM graph 401 is shown without anti-parallel

Dataln/RulesOut edges.

[0064] The downstream flow begins with the return supply fan Rfan node. Rfan is coupled to a duct segment Rduct

node which includes a Dataln edge for temperature sensor

Tra data.' Rduct is coupled to a mixing box Mixer node

which includes a Dataln edge for outdoor temperature

sensor Toa data. Mixer is coupled to a duct segment Mduct node which includes a Dataln edge for mixing temperature sensor T a data .· Mduct is coupled to two successive coil Hcoil and Ccoil nodes for heating and cooling. Ccoil is coupled to a duct segment Sduct node which includes a

Dataln edge for temperature .sensor Tsa data for

indicating the AHU' s output temperature and a Dataln edge for pressure sensor Psa data for indicating the AHU' s output pressure.

[0065] AHU 103 can be modeled as one HFM node at a higher level of the hierarchy with inputs for air and water, if the heated water supply system is modeled.

[0066] Each HFM node performs two functions: 1) a calculation of downstream FwdOut edge vector A _/+i and upstream RevOut edge vector state variables that it outputs to adjacent nodes based upon estimates derived from HVAC control system data {Dataln) and the node configuration, and 2) a calculation of one or more failure rules {RulesOut) as applied 'to the downstream

Fwdln edge vector A _j and upstream Revln edge vector 2? _; state variable parameter ranges based upon estimate parameter ranges derived from control system data

(Dataln) and the node configuration.

[0067] Each HFM node receives a downstream vector A _j and an upstream vector B . Depending on a particular node's functionality (behavior) , one or more downstream vector A _j and/or one or more upstream vector B _} state variables may change. State variables that change are reflected in the downstream A _j+] and upstream B _J+] vector state

variables output by that node. The downstream A _j+I and upstream B _J+ vector state variables are propagated to adjacent HFM nodes (FIG. 4) .

[0068] Each HFM node performs an estimation calculation.Estimations are different from building performance simulators since they lack detailed information about the components of the supervised system, the behavior of the control system and precise sensor data. Because of the lack of detailed information, an HFM node cannot perform full dynamic calculations. However, since HVAC control systems react slowly to environmental perturbations, steady state behavior is considered.

[0069] What is not neglected are frequent rule

activations due to oscillations in the control system or because too many control actions can reduce the lifetime of the components. It is the task of fault diagnosis to distinguish between different types of short term rule activations .

[0070] Node rule estimations are based on the downstream A _j and upstream B _j vector state variables, the affect the node has on one or more of the vector state variables, and sensor/control data input from the HVAC control system during runtime.

[0071] For the following example, only downstream vector A _j temperature parameter range Tmin and Tmax are

considered. Node Hcoil receives vector A ₅ having a temperature measured two HFM nodes upstream in Mixer by temperature sensor Tma . Hcoil receives HVAC control system (Dataln) data that modulates its heating coil inlet control valve Hcv. The maximum heating power from the BIM is stored as an Hcoil node specific configuration parameter .

[0072] FIG. 2 shows the problem of estimation in that there are many uncertainties, beginning with the

tolerance of the air temperature sensor Tma and the unknown air temperature distribution across the duct in which the sensor is mounted. The heated water temperature is not known nor is the heating water flow rate. It may be assumed that the maximum heating power can be ^' reached if the heating water inlet valve is 100% open and water pressure and temperature are at the maximum design values. Therefore, the Hcoil node air output temperature can best be estimated when the valve is closed, or a large range of possible values must be assumed.

[0073] For this reason, each node receives a downstream vector A _f and an upstream vector B _j . Each upstream and downstream vector parameter is expressed as a minimum and a maximum of an estimated parameter range. If a normal, Gaussian distribution is used for the probability density function of expected values to be in the range, a mean value ± standard deviation is used as range limits. In most cases, the shape of the distribution is not known and assumed range limits define the range.

[0074] The first step to derive a node's failure rule(s) are to list possible and probable faults. For example, temperature and pressure sensors outputs may drift or fail at one value. Valves and dampers may leak or fail in their last position. Filters and pipes may clog. Process controllers may malfunction. Design faults are also possible. For example, the cooling coil may be undersized and not compensate for an expected heat load. [ 0075] Any of these faults can cause several failures that can be detected internally or externally. External failures can be directly measured or felt by occupants. For example, a space temperature deviating from a controller setpoint. Internal failures refer to

situations where the control system compensates for a fault. For example, a leaking heating coil valve is compensated for by the control system increasing cooling. The user is not affected, but energy consumption is increased. The detection of internal failures is also important because of failure propagation. The closer to the fault a failure is detected, the easier it is to locate .

[ 0076 ] A failure rule is a conditioned inequality,

[0077 ] rule = {condition .exprl < expr2— threshold) ,

[ 0078 ] where exprl is a vector state variable parameter range minimum, exprl is a parameter range estimate maximum and threshold is a tolerance. If a rule output is true , a fault exists. Conditions are premises of control values .

[0079 ] AHU 201 has mixed air temperature Tma and supply air temperature Tsa control system sensor measurements. Node Hcoi l includes a modulating control valve Hcv and has a Da ta ln edge control input ^^ . Node Ccoi l includes a modulating control valve Ccv and has a Da ta ln edge control input n _cc . The building HVAC control system outputs a control variable in a range of 0 to 1 to modulate control valves. [0080] rulel = = 0 and u _cc = 0 : Tsa < Tma-ε^ j

[0081] Rulel states that if neither heating nor cooling is provided (u _hc = 0 and u _cc = θ) , the supply temperature Tsa should not be less than mixed air temperature Tma, with the combination of a mixed air temperature Tma sensor tolerance and the supply fan Sfan heat load expressed as threshold ε _] . If rulel is true, the cooling coil valve Ccv may be leaking and/or temperature sensors Tma and/or Tsa may have experienced a fault. If the heating coil valve Hcv is leaking, rulel would be true. The example shows that several faults can trigger the same rule. The opposite also holds: one fault can trigger several rules.

[0082] The above downstream direction temperature rule example shows that besides the control system inputs u _hc and u _cc for modulating control valves Hcv and Ccv, Tma and Tsa sensor values are required from nodes Mduct and

Sduct. Nodes Mduct and Sduct are separated by nodes

Hcoil, Cecil and Sfan. For modularity rulel (4) has to be evaluated in one of the five nodes. None of the five node types has both Tma and Tsa temperature sensor values.

[0083] The solution to needing both Tma and Tsa .

temperature sensor values is downstream and upstream propagation of sensor data via the downstream A _} and upstream B _j vector state variable parameter ranges.

[0084] Instead of a direct propagation, which results in several values propagated in parallel, data are

transformed by the nodes according to their physical behavior . [ 0085] For example, node Hcoil receives temperature

measurement Tma from upstream node Mduct. Node Hcoil

increases Tma according to the control variable u _hc and creates a heating coil temperature The. Node Hcoil

propagates The downstream to node Ccoil. Node Ccoil

receives temperature The and decreases The according to the control variable u _cc and creates a cooling coil

temperature Tec. Node Ccoil propagates Tec downstream to node Sfan. Node Sfan receives temperature Tec and

increases Tec by a predetermined amount due to fan motor heat load and creates a supply fan temperature Tsf. Node

Sfan propagates Tsf downstream to node Sduct . Node Sduct

receives temperature Tsf and applies rulel in a modified form using only local variables as

[ 0086] rule\ = M^ = 0 and u _cc = 0 :Tsa < Tsf— ε ^ ) · (5)

[ 0087 ] The threshold ε ₂ has changed because the heat load of node Sfan is considered in temperature Tsf. In

principle, rulel (5) can be generalized because the

inequality signals a fault for any combination of the

control variables _hc and u _cc . This results in

[0088] rule! = < Tsf - )■ (6)

[0089] Because propagation works upstream as well, rules equivalent to (5) and (6) can be evaluated in each of the five involved nodes. A failure decision depends on

several general principles .

[ 0090 ] One principle is a rule should be applied in the node where a fault may trigger it. Since this relation is not one-to-one, the same rule can be evaluated in several nodes. For rulel (5), the applicable nodes would be nodes Mduct, Ccoil and Sduct because a fault in either one would trigger the same rule in all three nodes. This redundancy has to be solved by the FDD which is executed in a node at the top of the hierarchy. The advantage of this concept is that nodes can be easily typed and reused for implementations.

[0091] Another principle is to evaluate a rule in the node in the flow that provides the most precise state variable value, typically a sensor value. In the example, this is node Sduct that receives temperature sensor data Tsa .

[0092] The problem of vector state variable parameter values is inherent tolerance. For example, in (4), the tolerance is observed by threshold £·, . The value of threshold £·, depends on the affect of several nodes. This decreases the modularity and reusability of node models.

[0093] To overcome these problems, embodiments calculate downstream vector A- and upstream vector B _j state variable parameter ranges within each node, and propagate the calculated downstream vector A _j+] and upstream vector

5· ₊₁ state variable parameter ranges to an adjacent node.

A node rule tolerance is a function of control system Dataln edge values received during runtime and state variable configuration parameters for the node obtained from design data of the supervised system. The tolerance is also the result of estimation uncertainties.

[0094] For example, the temperature sensor Tma value from node Mduct may have a tolerance of ±0.5°C. During runtime, 21 °C is measured and input (Dataln) to Mduct. Node Mduct propagates the vector A _t state variable

temperature range Tmin, Tmax as

[0095] Tma = 2 1 .0°C + 0.5° , (7)

[0096] Mduct FwdOut = A _i = (8)

pmin

[0097 ] to node Hcoil and propagates the vector B _} state variable temperature range Tmin, Tmax as

Tmax 21.5

Tmin 20.5

Qmax(

Qmin (

[ 0098 ] Mduct RevOut : 9 )

Hmax(

Hmin(

pmax(

pmin (

[ 0099] to node Mixer.

[00100 ] Node Hcoil has configured, predefined values for the maximum air temperature increase at maximum mass air flow rate and maximum heating water temperature, each

parameter having a tolerance. Because the mass air flow rate has not been measured by the HVAC control system, it may be assumed that the maximum air temperature increase

TincrMax = 20.0°C is independent of the air flow rate. During runtime, node Hcoil control valve Hcv may receive a

control variable u _hc - 0.5 (Dataln) . If the heating coil air temperature increase is proportional to the heating water flow rate, the control variable indicates a heating water flow between 40 % and 60 % considering nonlinearity of the valve. Therefore, node Hcoil's maximum air temperature

increase TincrMax estimate is

[00101] ThcMax - TmaMax + 0.6(TincrMax) ( 10 )

[00102] ThcMax = 21.5° + 12.0° (11)

[00103] ThcMax = 33.5° (12)

[00104] Using (10), node Hcoil 's minimum air temperature increase ThcMin estimate is calculated as 28.5°C. This is a local calculation in one node without knowledge about

other nodes. Node Hcoil propagates the vector A ₆ state

variable temperature range T in, Tmax as

[00105] Hcoil FwdOut = A _b ( 13 )

pmin

[00106] FIG. 5 shows the downstream propagation of the air temperature range Tmin, Tmax starting at node Mduct and passing through Hcoil, Ccoil and Sfan and ending at

Sduct .

[00107] The resulting large tolerances make the state variables useless. This allows the use of the more general rule2 (6) instead of rulel (5) . For the case of heating valve Hcv closed u _hc - 0 , the tolerance shrinks to that of temperature sensor Tma and if the cooling valve Ccv is closed u _cc = 0 , (6) includes the case (4) .

[00108] Using the vector state variable parameter range representation, failure rules may be generalized. This is an important contribution to node reusability. For example, node Sduct receives vector A _K having a

temperature range Tmin, Tmax and a Dataln edge temperature sensor value Tsa which is transformed into a temperature tolerance range .

[00109] As a first approximation it can be assumed that no fault can be deduced if the vector state variable temperature range and Tsa temperature tolerance range overlap, because there is a probability that the real air temperature is within the union of both ranges. A fault is presumed if the vector state variable temperature range and Tsa temperature tolerance range do not overlap.

[00110] FIG. 6 shows five situations for node Sduct

receiving vector A having a temperature range Tmin, Tmax and comparing it to five different calculated estimate temperature ranges for temperature sensor Tsa . Cases 1 and 5 are for two different failure rule activations, case 3 is within limits and therefore no fault, and cases 2 and 4 have a probability of representing a fault. [00111] As a refinement, a continuous failure value can be calculated in addition to the result being true or false.

In its simplest form the gap between a parameter range,

and its corresponding calculated estimate parameter range can be taken as a measure of the size of the failure if

true. FIG. 6 shows two error calculations shown as double arrows for cases 1 and 5. The reason is if a rule

evaluates to true, there is still a probability that no

failure has occurred because the tolerance range or a

threshold may have been chosen too small. The larger the rule value, the lower the probability will be, or

expressed differently, the higher the probability of a

failure. Therefore, this rule value can be used for

failure analysis in addition to the values true and

false .

[00112] The following equations represent the solution for the above example .

[00113] rule 3 = max(0, TsaMin - Fwdln Tmax) and (14)

[00114] rule4 = max(0, FwdIn Tmin -TsaMax) , (15)

[00115] where TsaMin is an estimated range minimum and

TsaMax is an estimated range maximum. As long as an

overlap exists, rul e3 and rul e4 are zero. For case 1,

rul e3 > 0. For case 5, rule 4 > 0. Cases 1 and 5 indicate failures .

[00116] A more general solution is the relation between

two temperatures ranges Tl and T2,

[00117] rule l = max( , T ₂ Min - T _l Max) , and (16) [00118] rule 2 - max(0, T^Min- T ₂ Max) . (17)

[00119] An extension to (16) and (17) would be

normalization. (16) and (17) can also be applied to

another temperature range pair TsaSens and TsaSet to check if the temperature control is working properly.

Temperature is used an example. Any state variable

parameter can be used as rules for failure checking.

[00120] The above shows that nodes of the HFM graph can

perform estimation and rule evaluation without knowing

anything about the rest of the graph besides knowing its neighboring nodes for the communication of flow vectors.

This is a case for object orientation with information

hiding. For the purpose of generating distributed

systems, multi-agent technologies can also be applied

without problems.

[00121] Not all problems can be solved locally in the

nodes at the granularity level that has been shown. This generally is not a problem since a graph is hierarchical and composed nodes are possible as AHU 401 in FIG. 4. AHU

401 is a composed HFM node.

[00122] There are also nodes which are not part of the

flow such as a controller node. Controllers evaluate

different rule types. For example, if a heating and

cooling coil control valve control variables are greater than zero at the same time, one or both controllers have sustained a fault.

[00123] The advantage of object orientation is the use of types and the derivation of subtypes by inheritance. One type can also produce many instances at runtime. For the purpose of automatic generation of an executable FDD system from a BIM, a library of HFM node types has to be provided. The goal should be to keep this library small to reduce software design and maintenance cost.

[00124] AHU 401 shows five different node types. The node examples are single air flow input and output nodes. The Mixer node does not model the flow interfaces to the outdoor environment. The Hcoil and Ccoil nodes do not model their associated heating and cooling water flows. Such simplifications are appropriate if nothing is known about omitted parts. However, modeling the total system improves estimation and rule evaluation.

[00125] FIG. 1 _. shows that the AHU 103 serves several VAV 105 zones. This requires Branch nodes with one air flow input and several outputs and Joint nodes with several air flow inputs and one output. The VAVs 105 are modeled as complex nodes with several components encapsulated. The number of. VAV types remains small enough for a node library .

[00126] FIG. 7A shows a sensor duct node that includes Dataln edges Xsens and XSet . XSet is a controller setpoint and Xsens is a sensor variable. Xsens is externally controlled to be within a range of XSet. Rule3 (14) tests if this is the case. Estimators toll and tol2 transform runtime Dataln XSens and XSet inputs into XSens and XSet tolerance configuration ranges. Rulel (16), rule2 (17) and rule3 (14) apply.

[00127] The sensor duct node is general and can be applied to temperature, flow, humidity and pressure sensor situations. One or more sensors can be modeled in the same duct, representing controlled or uncontrolled variables. The same node type can also be modified for

water flow nodes. FIG. 7A does not show flow connections from the inputs directly to the outputs for the variables that are not sensed. These variables do no change in

short duct sections and are directly propagated.

[00128] FIG. 7B shows a temperature sensor duct node that includes Dataln edge Tsens. The Tsens temperature

tolerance range is

[00129] TsensMctx = Tsens + tolerance , and (18)

[00130] Tsen Min = Tsens - tolerance . (19)

[00131] Rulel (16) and rule2 (17) compare the Tsens

tolerance range TsensMin. TsensMax with . and B _} vector

temperature ranges Tmin, Tmax .

[00132] FIG. 7C shows a flow controlled fan node that

includes Dataln edge Q (flow sensor measurement) . A fan

node is an estimation function. The air temperature

typically increases in the order of 1°C from Fwdln to

FwdOut due to heat load. Since the increase includes a

tolerance, the FwdOut A _J+[ temperature range Tmin, T ax

expands and is wider than the Fwdln A _} temperature range

Tmin, Tmax . In the upstream direction, the temperature

range Tmin, Tmax decreases from Revln B _j to RevOut B _J+I by

the same amount.

[00133] If the fan is a constant flow rate type, it can

propagate the flow rate downstream and upstream. FIG. 7D shows a pressure controlled fan node that includes Dataln

edge P (pressure sensor measurement) . For fans of a

constant pressure type, the pressure can be propagated downstream, but not upstream. Pressure controlled fans

vary air flow rate. Therefore, the air temperature

increase ( FwdOu t A _J+l temperature range Tmin, Tmax ) changes as a function of rate. Modern fans reduce electric power proportional to the air flow rate and thus keep the

temperature increase constant.

[00134] The flow controlled fan node (FIG. 7C) propagated temperature ranges:

[00135] FwdOut T x = Fwdln Tmax + dTfmax , (20)

[00136] where dTfmax = maximun fan heat load ,

[00137 ] RevO t TMax = RevlnTMax - dTfmin , (21)

[00138] where dTfmin—minimum fan heat load ,

[00139] FwdOut Tmin = FwdIn Tmin + dTfmin , and (22)

[00140] RevOut Tmin = RevIn Tmin - dTfmax . (23)

[00141] Propagated flow ranges:

[00142] FwdOut Qmax = Qmax , (24)

[00143] RevOut Qmax - Qmax , (25)

[00144] FwdOut Qmin = Qmin , and (26)

[00145] RevOut Qmin = Qmin . (27)

[00146] Ru l el (16) and rul e2 (17) compare estimated

temperature and flow with propagated values. The pressure controlled fan node (FIG. 7D) propagated temperature

range :

[001 7] FwdOut Tmax - Fwdln Tmax + dTfmax , (28)

[00148] where dTfmax = maximun fan heat load ,

[00149] RevOut Tmax = Revln Tmax - dTfmin , (29)

[00150] where dTfmin = minimum fan heat load ,

[00151 ] FwdOut Tmin = Fwdln Tmin + dTfmin , and (30)

[00152 ] RevOut Tmin = RevJn Tmin— dTfmax . (31)

[00153] Propagated pressure range:

[00154] FwdOut P ax = Pmax , ( 32 )

[00155] RevOut Pmax - Pmax , (33)

[00156] FwdOut Pmin = Pmin , and ( 34 )

[00157] RevOut Pmin = Pmin . (35)

[00158] FIG. 7E shows a coil node that includes Dataln

edge ctrlin. Coil nodes for heating or cooling have

complex physical models. The number of unknown parameters and state variables is large. Therefore, it is necessary to create several typical coil node types in the library or one type with different selectable estimators.

[00159] FIG. 7F shows a thermostat node that includes

Dataln edges Tsens and TSet. TSet is the thermostat setpoint and Tsens is the temperature measurement process variable. The propagated temperature range is

[00160] FwdOut T x = Tsens + tolerance , (36)

[00161] FwdOut Tmin = Tsens— tolerance . (37)

[00162] Rulel (16) and rule2 (17) compare a node

temperature tolerance range with a propagated (Fwdln,

Revln) temperature range. Rule3 (14) compares a setpoint

Tset estimate with an estimated temperature measurement.

[00163] FIG. 7G shows a mixer box node that includes

Datain edge Toutdoor Tsens/Damper Ctrls. A physical

mixing box includes a temperature sensor Toa for

measuring outdoor air and changes supply air temperature measured by another temperature sensor Tma by mixing

outdoor air with return air measured by another

temperature sensor Tra . The mixing ratio is controlled by three modulating dampers (FIG. 2) . The supply flow rate

equals the return flow rate if no outside air is

admitted. But it cannot be assumed that the mixing rate

is proportional to the damper control signal from ^' an AHU controller. There are two extremes when the outdoor air

damper is either fully closed or fully open that the

mixing rate is precisely known. The largest deviation due to nonlinearity can be assumed at 50% open. This relation is considered in the mixing function when estimating the mixed air temperature and humidity.

[00164] The propagated temperature range is

[00165] FwdOut Tmox = Tsens + tolerance , and (38)

[00166] FwdOut Tmin— Tsens— tolerance (39) [00167] Est2 estimation for temperature range T for Tma

using damper opening percentage and Toa estimation.

[00168] Rulel (16) and rule2 (17) are comparisons of the temperature tolerance range with propagated temperature

range. Rule3 (14) is a comparison of estimated Tma using the damper openings and thermal dynamics to estimated

measurement of Tma.

[00169] FIG. 7H shows a two-way branch node. A duct branch propagates air temperature, flow, pressure and humidity, and allows reverse calculation of air flow rate. Air flow rate is measured in all VAVs . The sum of all measured

values is the air flow rate in the input of the branch as a RevOut value.

[00170] Estimation for temperature and flow range:

[00171] FwdOutX Tmax = Fwdln Tmax + dTfmax , (40)

[00172] where dTfmax = change through branch ,

[00173]

RevOut Tmax = Max (Revfn\ Tmax - dTfmin, Revlnl Tmax— dTfmin) , (41)

[00174] where dTfmin = change through branch ,

[00175] FwdOutl Tmin - Fwdln Tmin + dTfmin , (42)

[00176] RevOut Tmin = Min (Revlnl Tmin - dTfmax, Revlnl Tmin - dTfmax)

(43)

[00177] FwdOutl Tmax = Fwdln Tmax + dTfmax , (44)

[00178] FwdOutl Tmin = Fwdln Tmin + dTfmin . (45) [00179] Rulel (16), rule2 (17) and rule3 (14) compare

temperature tolerance range with propagated temperature

range .

[00180] FIG. 71 shows a reheat VAV node that includes

Dataln edges Xsens and XSet-. XSet is a controller

setpoint and Xsens is a sensor variable. A reheat VAV is a complex node. If equipped with sensors for air flow

rate and distribution temperature, an air flow rate can

be estimated using the Fwdin edge propagated pressure

range and the damper control value. Using the input air

temperature, the measured flow and for example, the

electric power value of the reheat coil, the distribution temperature can be estimated. Both estimations can be

evaluated with the sensor values in failure rules using

(16) and (17). In addition, if a setpoint is defined, the compliance can be evaluated as well.

[00181] Estimation for temperature range is

[00182] FwdO t Tmox = Tsens + tolerance , and ( 6)

[00183] FwdOut Tmin = Tsens - tolerance . (47)

[00184] Rulel (16) and rule2 (17) compare estimated

temperature ranges with propagated temperature ranges.

Rule3 (14) compares estimated setpoint Tset with

estimated measurement Tsens.

[00185] Space that is serviced with heating or cooling

from one VAV and has a temperature sensor to control the

VAV can propagate ranges downstream and upstream. As a

minimum, it can evaluate if the air temperature is at

setpoint. If the heat load and loss in the space is

known, especially in closed rooms, estimations of the necessary heating or cooling power can be compared with the distributed air parameters of the VAV as a failure rule. This may require the outdoor temperature as a data input .

[00186] The return air from spaces is collected in a joint duct. Failure rules may be applied if the temperature of the joined air can be estimated. The air temperature in each space is known with tolerances at the location of the sensor for the control of the VAVs . At the air intakes the air temperature will be different, adding to the tolerance. If the air flow rate of the return air of each space would be known, temperature ranges for the joined air can be calculated. In open plan offices individual flow rates are not known. The upper limit of the range can not be higher than the largest measured sensor value, with the lower limit not less than the smallest. The FwdOut edge temperature range can be used to detect for example a large fault of the return air sensor .

[00187] HFM graphs specify the physical structures and fault detection functions. Embodiments specify a

hierarchical structure of autonomous communicating processes that can be interpreted as agents. The HFM nodes can be directly mapped into, for example,

Specification and Description Language (SDL) processes. The advantage of SDL models is that a model can be automatically translated into C code and compiled into executable prototype systems. SDL prototype experiments show that the introduced faults lead to positive failure rule outputs that are collected in files.

[00188] Embodiments comprise an engineering phase and a runtime phase. FIG. 8 shows an HFM building FDD system 801. The HFM building FDD system 801 comprises an interface 803 configured to access and import BIM/IFC files, a Graphic User Interface (GUI) 805 configured to allow additional data to be input during graph assembly and view linkage information among different components and enter and configure missing parameters, conditions, and switches not found in the BIM, and a compiler 807 to compose together the file data with the additional engineering information.

[00189] The benefit of HFM is the engineering efficiency it brings to HFM based FDD systems. Due to the modularity of HFM, it can be automatically composed from the BIM and graph models can be effectively compiled into executable FDD systems. IFC may be used to derive an HFM model since IFC is one of the most commonly used BIM formats.

[00190] IFC provides a set of definitions for all object element types encountered in HVAC mechanical and control system and a text-based structure for storing those definitions in a data file. It comprises two major parts: IfcElement and IfcPort. An element can be any component that can be connected to neighboring elements, using one or many ports. The IFC elements include flow segments (duct) , flow fittings (duct joint) , moving devices

(fans), flow controllers, etc. Every HFM node can be mapped from IFC elements.

[00191] The IFC element objects have defined the basic properties and attributes which are used for HFM node graphs. FIG. 9 shows a list mapping relationship between IFC HVAC elements and HFM node types. The inheritance of an HVAC node type is realized through the associated IFC type definition of each element. There are no direct

35 mappings for node types such as Mixer, Reheat VAV and they are composites of basic IFC elements.

[00192] The connectivity model for HFM can be derived through IfcPort. A port is a point in the network at which elements are connected to each other. An IfcPort is associated with an IfcElement, it belongs to, through the objectified relationship IfcRelConnectsPortToElement . Therefore, the linkage information for HFM can be obtained through searching IFC model for the IFC element objects, IFC port objects, and

IfcRelConnectsPortToElement objects. FIG. 9 shows the correspondence between IFC types and HFM nodes.

[00193] HFM reduces the engineering effort required to configure FDD systems for different HVAC systems. HFM is the bridge between the BIM and the compiled graph model based executable FDD systems .

[00194] The compiler 807 is coupled to an HFM engine 809 that is configured to generated an HFM model in an extensible Markup Language (XML) format, the HFM model is input to an FDD generator 811 configured to populate the identified HFM node types with functional HFM graph nodes from an HFM node library 813.

[00195] The HFM graph derived offline can be loaded into a runtime FDD system to instantiate an FDD engine 815 for the specific building HVAC control system and configured to execute the resultant HFM graph inputting HVAC control system process and control variables from an input interface 817. During runtime, the real-time HVAC control system data is input to an FDD engine 819 configured to analyze the rules employed in the HFM graph. [00196] HFM makes fault detection for HVAC system realistic. To save the engineering effort for configuring HFM based FDD for each specific building, object

orientation is used for modeling and also for

implementation. Objects in the model represent real components with functions to capture the correct behavior and rules for detecting faults. Object models are provided in a node library and are assembled as graphs by signal path as real components are connected by ducts and pipes. Objects can be hierarchically composed and decomposed.

[00197] HFM can have many different node types. Using inheritance, more node types does not mean more effort to extend the library.

[00198] Whether the FDD is centrally hosted by a building management system or distributed in digital process controllers, it receives runtime control signals and sensor data from the respective HVAC control system. For each HFM node, based on received measurements and flow information from upstream and downstream nodes, HFM makes state estimation, propagates resulting vector parameter ranges to adjacent nodes, and detects faults accordingly. The faults are also propagated up in the component hierarchy and a diagnosis is made.

[00199] During the engineering phase, the GUI 805 is used to edit HFM nodes and graphs from BIMs . It first

identifies all the HVAC node objects from IFC and retrieves the useful attributes to create node estimation models. It also creates the connectivity among different nodes from IFC connectivity data. For those missing parameters, conditions, and switches not found in the BIM, the tool provides an interface for the user to enter data. Otherwise, default settings are used.

[00200] This information is compiled 807 together and an HFM model is generated in XML format. The XML based HFM graph is loaded into an HVAC FDD system, which compiles the graph model into an FDD engine 815 based on its runtime compiling capability and an existing object library for HFM nodes. The FDD engine 815 can be embedded in the existing Building Management System (BMS) to perform fault detection and diagnosis using the rules modeled in each node and the propagation of faults.

[00201] The FDD engine 815 may be implemented using

Specification Description Language (SDL) modeling tools. SDL specifies a hierarchical structure of autonomous communicating processes that can be interpreted as agents. Flow nodes can be directly mapped into SDL processes. The advantage of SDL models is that a model can be automatically translated into C-code and compiled into executable prototype systems.

[00202] One or more embodiments of the present invention have been described. Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the invention. Accordingly, other embodiments are within the scope of the following claims .