Cross-Reference to Related Applications

[0001] The present application claims priority from Australian Provisional Patent Application No 2021903430 filed on 26 October 2021, the contents of which are incorporated herein by reference in their entirety.

Technical Field

[0002] The present disclosure relates to a method and system of modelling and inspecting components of a bicycle. In some particular applications, this includes computer modelling and inspection of fibre-reinforced plastic bicycle frames for damage using non-destructive testing techniques.

Background

[0003] Carbon fibre bicycle frames and components can experience catastrophic failures resulting in severe injury or death. This can result from contextual factors such as inappropriate usage of the bicycle and/or reduced integrity due to damage. Damage is not always visible to the naked eye due to size or type of damage, or it may be on the interior of the frame or component. It can be caused via physical impacts such as crashes, or can even be present due to manufacturing defects such as delamination and resin-rich areas. Damage is also difficult to quantify - e.g. what size, type of damage and location of damage will substantially increases risk of catastrophic failures? Damage to carbon fibre bicycle frames and components can result in sudden, unexpected failure (in contrast to failure modes from metals and metal alloys).

[0004] Damage identification and interpretation of non-metal materials such as carbon fibre reinforced polymers (CFRP) is a highly technical process requiring appropriate training, technology, and experience. Non-destructive testing (NDT) of composite materials is heavily regulated within the defence and aerospace sectors. However, no such regulation exists within the bicycle industry. [0005] The bicycle inspection process, as is currently offered, lacks a standardised inspection method/s or any governance and oversight of the application of such method/s. Whether formally or informally, businesses or professionals at any level within the industry who offer advice relating to the severity of damage/defects expose themselves, their business, their customers and stakeholders through a lack of inspection rigour, quality control and diagnostic intelligence.

[0006] It is desirable to have a standardised technical method and system to inspect and assess the condition of one or more components of a bicycle and provide an objective output of roadworthiness and actionable items to improve roadworthiness if necessary (e.g. suitable repair options) and to monitor damage with follow-up inspection schedules. This is important for safety of the rider as well as management of servicing and repair of the bicycle to provide longevity for the components.

[0007] Any discussion of documents, acts, materials, devices, articles or the like which has been included in the present specification is not to be taken as an admission that any or all of these matters form part of the prior art base or were common general knowledge in the field relevant to the present disclosure as it existed before the priority date of each claim of this application.

[0008] Throughout this specification the word "comprise", or variations such as "comprises" or "comprising", will be understood to imply the inclusion of a stated element, integer or step, or group of elements, integers or steps, but not the exclusion of any other element, integer or step, or group of elements, integers or steps.

Summary

[0009] There is disclosed an inspection method comprising:

- receiving a plurality of contextual factors based on usage and history of a bicycle;

- scanning the bicycle (e.g. visually and via NDT methods) to detect potential damage at a plurality of sites at the bicycle; - categorising and sizing the potential damage, at each of the plurality of sites, to generate a plurality of damage factors;

- comparing the plurality of contextual factors and plurality of damage factors with a pre-existing data set from a database to form a comparison;

- determining roadworthiness of the bicycle based on the comparison; and

- displaying, at a user interface, an indication of the roadworthiness of the bicycle. This may also include listing contextual and damage factors of concern, suggesting suitable repair options, and indicating likely costs of repair.

[0010] Examples of the present disclosure can be used for safety inspections of carbon fibre bicycle frames and components using non-destructive testing (NDT) techniques. This may include the identification and assessment of damage and/or defects uncovered during the inspection process. In some examples, this may include providing a risk assessment.

[0011] In some examples, the inspection method further comprises determining potential corrective action based on the comparison. The method also includes displaying an indication of the potential corrective action for the bicycle.

[0012] In some examples, the method further includes updating the pre-existing data set based on comparing the pre-existing data set with the plurality of contextual factors and the plurality of damage factors of the bicycle.

[0013] There is also disclosed a computer program that, when executed by a computer, causes the computer to perform the computer-implemented method described above.

[0014] There is also disclosed a system comprising: a scanner to scan a bicycle for potential damage; at least one processing device to perform the inspection method described above; and a display for outputting the indication of roadworthiness and/or potential corrective action.

Brief Description of Drawings

[0015] Fig. 1 illustrates a system to model and inspect a bicycle; [0016] Fig. 2 is a flow diagram of a method of inspecting a bicycle;

[0017] Fig. 3 illustrate the concept of modelling and inspecting a bicycle;

[0018] Fig. 4 is a screenshot of a computer application for inspecting a bicycle;

[0019] Fig. 5 illustrates a user interface of the application to enter contextual factors;

[0020] Fig. 6 illustrates a user interface showing detected bicycle damage and respective damage factors;

[0021] Fig. 7 illustrates a user interface showing an indication of roadworthiness and corrective action;

[0022] Fig. 8 is a visualisation at a user interface of the bicycle and detected damage in an interactive window;

[0023] Fig. 9 is an example table of contextual factors and the relationship with roadworthiness as part of pre-existing data;

[0024] Fig. 10 is an example table of damage factors for a bicycle frame and the relationship with roadworthiness as part of pre-existing data;

[0025] Fig. 11 is an example table of damage factors for bicycle components and the relationship with roadworthiness as part of pre-existing data;

[0026] Fig. 12 is a table showing risk ratings based on risk score;

[0027] Fig. 13 is an alternative screenshot of a computer application for inspecting a bicycle;

[0028] Fig. 14 illustrates an user interface showing detected bicycle damage and respective damage factors: and

[0029] Fig. 15 is a schematic example of a processing device. Description of Embodiments

[0030] Overview of the system

[0031] Fig. 1 illustrates an example of a system 2 to model and inspect a bicycle 1. This includes a scanner 51 to scan a plurality of sites (5) at the bicycle to detect potential damage 3. The scanner 51 may include non-destructive scanners such as optical cameras, ultrasonic sensors, X-ray scanning systems, tap hammers etc.

[0032] The system 2 also includes at least one computing device 53 that includes at least one processor 55, and a user interface such as a display 57. The display 57 outputs an indication of roadworthiness and/or potential corrective action determined by the method described below. In some examples, the computing device 53 is a computer (including desktop computer, laptop computer, tablet computer, mobile communication device, smartphone, specific purpose computing device etc.). In some examples, the computing device 53 is also in communication with at least one other computing device 61, 63, and data storage 16. This can include direct communication, or over a communications network 59.

[0033] Overview of the method

[0034] An example of an inspection method 100 performed by the system 2 will now be described generally with reference to Fig. 2. The method 100 includes receiving 110 a plurality of contextual factors based on usage and history of a bicycle 1. This will be discussed in further detail below and may include contextual factors such as type of use, age of components of the bicycle, rider weight, time between servicing, etc.

[0035] The method also includes scanning 120 the bicycle 1 to detect potential damage 3 at a plurality of sites 5 at the bicycle 1. This may include scanning 120 for damage such as cracks, delamination, deformation, etc..

[0036] The method 100 further includes categorising the potential damage 3 (if found at the plurality of sites 5) to generate a respective plurality of damage factors 13. Categorisation may include categorizing the size, type, age, and other characteristics of the damage 3. This enables categorization and generation of damage factors that can be used as objective measures for the model to compare and assess the state of the bicycle 1. [0037] The method further includes comparing 140 the plurality of contextual factors 11 and plurality of damage factors 13 with a pre-existing data set 15 from a database 16. The preexisting data set 15 may include data to model the performance, lifespan, wear and fatigue, maintenance, and other characteristics of the bicycle 1 and bicycle components. As described in further detail below, the pre-existing data set 15 and model may be updated with machinelearning models.

[0038] The method 100 further includes determining 150 roadworthiness 19 of the bicycle 1 based on the comparison. An indication of the roadworthiness 19 of the bicycle is then displayed 160 at a user interface 21 for a user.

[0039] In some examples, the method 100 also includes determining 151 potential corrective action 25 based on the comparison 17. This can include replacement, repair, maintain, inspect, or other action. This may also include providing information on qualified repairers, suppliers, manufacturers, replacement or substitute parts, etc.

[0040] Examples of the system 2 and method 100 aims to provide standardised nondestructive testing for the cycling industry, particularly for CFRP components. This includes applications that apply statistical models that in conjunction with contextual factors 11 and damage factors 13, provide an effective and objective report on the state of the bicycle. Importantly, such reports can be produced objectively with results that are standardised and comparable with reports generated by other technicians using the system 2 and method 100. This above described process and concept 71 is also illustrated in Fig. 3.

[0041] Additional details of the system will be described in further detail followed by an example of the method 100 implemented, at least in part, on an application running on the computer 53.

[0042] The bicycle components and context of the problems

[0043] Current customer-centric approaches to carbon fibre bicycle and component inspection focus primarily on the identification and assessment of observable (visual) damage only - and often by inexperienced and unqualified bicycle mechanics. Relatively few practitioners (mechanics, repairers etc.) in Australia adopt NDT techniques that help identify underlying/invisible damage or defects. Outside of NDT, no solution currently exists that incorporates contextual information related to the owner, bike history, usage and care/maintenance. Such contextual factors may only present an elevated risk of failure when combined with other objective/contextual factors or when a certain threshold is reached.

[0044] The bicycle 1 includes a frame 7 that can be made of carbon reinforced plastic (CFRP), a mix of carbon fibre and other fibres reinforced with plastic, or other fibre embedded in plastics in a composite structure. In some examples, portions of the frame 7 are made of CFRP with other component materials embedded or joined therein. For example, mounts, fastener receiving portions, or other high-wear portions, made of metals or metal alloys may be molded in the frame 7.

[0045] The frame 7 can include various portions (such as sites 5), such as:

- head tube;

- top tube;

- down tube;

- seat tube;

- seat stay;

- suspension linkages; and

- chain stay.

[0046] One or more of these sites 5 of the frame 7 may advantageously be formed of composite materials such as CFRP for high strength to weight ratio.

[0047] The bicycle may also include other components that are not part of the frame 7, but also made of composite materials. This may include one or more of: fork; steerer; wheel rim;

- wheel hub;

- handlebar; crank; chainring; seatpost; and stem.

[0048] Scanner 51

[0049] In some examples scanning is performed by, or aided with, a scanner 51 configured to detect damage 3. Preferably, the scanner 51 utilises non-destructive testing techniques to detect damage. In one example, this may include an optical scanner (e.g. camera) or other measurement device to determine the size, location, and other characteristics, of visible damage such as cracks, delamination, breakages, etc. This may include optical or digital magnification to obtain clarity on the visible damage. In some examples, this can include measurement means, to aid in determining the size of the damage. Such measurement means may include an optical rangefinder, laser rangefinder (including LiDAR), gyroscopes, magnetometers, accelerometers, to provide information to enable determination of the size of damage. In some examples, the scanner 51 may be coupled with light sources to provide contrast and clarity on any potential damage. This can include using particular wavelengths of light (including visible and invisible wavelengths). In some examples, scanning 120 of the bicycle includes a technician’s input in operating the scanner 51, and/or interpreting the results, images, or other data from the scanner 51 to determine the nature of the damage.

[0050] In some examples, the scanner 51 includes an X-ray system to obtain one or more X- ray images of the bicycle. This may be particularly advantageous in detecting damage beneath the surface of the frame. For example, the damage may be internal, shielded by other opaque components and/or under paint or other coating on the bicycle.

[0051] In some examples, the scanner 51 includes an ultrasonic sensor system to perform ultrasonic inspection. This utilizes sound waves to determine inconsistencies in the material that are indicative of damage to the bicycle. Additional scanners operating with nondestructive techniques may be used, such as: thermography inspection; liquid-penetrant; radiographic testing; electromagnetic testing; eddy current testing; vibrational analysis, acoustic testing using a hammer, etc.

[0052] Computer 53, 61

[0053] The computer 53 may be a desktop computer, laptop computer, single purpose computing device, tablet computer, computer terminal, smart phone or other communication device. This can include a processing device 55, display 57, and a data store/memory. The computer 53 may perform, at least in part, the method 100 described herein. In some examples, the computer and associated data store contains a computer program (i.e. application) to perform the method 100. In other examples, the application is a web application hosted on the server 63, whereby the computer 53 acts as a terminal to provide a user interface with the technician.

[0054] In some examples, the computer 53 is in wired or wireless communication with the scanner 51 to receive scans, or associated scanner data. In further examples, the technician may assist in communicating data between the scanner 51 and the computer 53. For example, the technician may physically take a memory card, print-out, or other storage medium from the scanner 51 and transfer this to the computer 53. In other examples, the technician may read data from the scanner 51 and manually input the data to the computer 53.

[0055] In some examples, the computer 53 and application may be operative to send inspection reports or other information to a client’s computer 61. This may include email, short message service, application based messages/notification, etc.

[0056] Server and associated database [0057] The server 63 includes a processing device and associated data store 16. The server may be operated by an entity to provide subscription service to technicians at the computers 53. An advantage is that the entity may receive data from multiple sources (i.e. different technicians), which can provide a larger data set for training and the machine learning model. The server 63 is associated with a data store 16 that can hold historical records, pre-existing data sets and machine learning models 15. In some examples, the data store 16 also stores the computer program to perform the method 100.

[0058] In some examples, the server 63 performs the role of providing the computers 53 with pre-existing data sets. This may include sending updates of the pre-existing data sets. Thus the server may also receive data from actual inspections from the computers that can be used fortraining and updating the data sets. This may include the server 63 performing at least part of the machine-learning functions (described in further detail below).

[0059] In other examples, as noted above, the server 63 runs the application to perform the method 100 and communicates with the computers 53 that act as terminals.

[0060] It is to be appreciated that the server 63 may be a distributed server and/or a cloudbased server, and that the server may be either physical, virtual, or containerised. This may include providing the application as a web-based application to perform the method.

[0061] Example of an application performing an example of the method 100

[0062] Fig. 4 illustrates a screenshot 75 of an application user interface 21 (at the display of the computer) from an application performing the method 100 on the computer 53 (as a computer-implemented method). The steps of performing the method 100 of this example will be described in detail below.

[0063] Contextual factors 11

[0064] The application includes a model in which inspection inputs can be provided, and an estimate of overall risk of issues can be provided, along with some recommendations. In some examples of this model, various inputs about the level of risk associated with various types of contextual factors related to bicycle damage are received. These relate to contextual factors, that can include: - bicycle manufacturer;

- bicycle model;

- bicycle type (e.g. Casual or Serious/Racing on roads, cycleways, off-road, downhill mountain biking, enduro or jumping);

- bicycle usage type (e.g. Casual or Serious/Racing on roads, cycleways, off-road, downhill mountain biking, enduro or jumping);

- usage appropriateness (is the bicycle being used for its intended purpose);

- bicycle age (difference in years between the current year and the model year of the bicycle);

- rider weight;

- period since last bicycle service; and

- history, or unknown history, and duration of history of the bicycle.

[0065] These contextual factors may be entered into the user interface of the application as illustrated in Fig. 5.

[0066] These contextual factors will be entered into the model where it can be determined whether these factors can have an effect and/or are otherwise correlated with roadworthiness, damage and associated rectification steps. This relationship is included as part of the model in the pre-existing data set 15 in the database 16.

[0067] A simplified example of contextual factors 11 and inter-relationship with roadworthiness is shown in Table 1 in Fig. 9. As will be discussed in other examples below, the relationship may be trained and developed by machine -learning methods.

[0068] Damage factors 13 [0069] The damage factors 13 are further objective inputs that can be derived from potential damage 3 that are scanned by the scanner 51. For example, a scanner may scan multiple images of the bicycle at various locations (e.g. sites 5 of the bicycle noted above). These multiple images are then assessed to identify any potential damage 3 that is then categorised to the type of damage, as well as particular characteristics such as size of the damage. In some examples, the damage factors may include objective inputs from a technician. In some alternatives, the objective inputs from a technician may be performed by the human eye with, or without, the aid of a scanner.

[0070] An example of potential damage 3 that has been categorised to damage factors 13 for the application and method is illustrated in Fig. 6. This can range from “none” (i.e. no damage 14) to cosmetic, delamination 18, crack, as an illustrative example. The categorisation also includes the size 22 (length, depth, width) of that respective damage.

[0071] Examples of damage may refer to any irregularity in the carbon fibre reinforced plastic material and is likely not intended to be present by the original manufacturer e.g. cosmetic damage (including scratches and chips), cracks, visible impacts, delamination, disbands, fibre breakages, fibre misalignments, improper fibre splicing, inclusions, porosity, resin micro-cracking, resin-rich areas, unbonds, and voids.

[0072] The damage factors are relevant to the model as these factors have an effect and/or otherwise related to the assessment of roadworthiness and further steps to rectify and correct issues with the bicycle 1. This relationship is included as part of the model in the pre-existing data set 15 in the database 16.

[0073] A simplified example of damage factors 13 on the frame 7 of the bicycle and interrelationship with roadworthiness is shown in Table 2 in Fig. 10. A simplified example of damage factors 13 on other (non-frame) components of the bicycle 1 and inter-relationship with roadworthiness is shown in Table 3 in Fig. 11. Like contextual factors, this relationship of damage factors 13 in the pre-existing data set may be trained and further developed by machine-learning methods.

[0074] Comparing the factors with pre-existing data set and determining roadworthiness [0075] The method further includes comparing 140 the contextual factors 11 and damage factors 13 with pre-existing data set(2) 15 from the database 16. In one example, the preexisting data set 15 includes an association between the factors 11, 13 and a corresponding individual risk aspect.

[0076] Referring to Fig. 9, the contextual factor “Usage type” can includes “Roads & Cycleways: Serious or Racing” which is associated with pre-existing data that there should be “Caution” and has a risk score of “1”. An alternative “Usage type” includes “Off-road: Casual” that is associated with “Typical risk” and a risk score of “0”.

[0077] Referring to Fig. 10, the damage factor “Head Tube” that has “Crack or impact greater than 5mm” is associated with pre-existing data that is should be an “Extreme risk” and has a risk score of “10”. An alternative damage factor for the “Head Tube” that has “No damage” is associated with a risk aspect of “Typical risk” and with a risk score of “0”.

[0078] The plurality of contextual factors 11 and damage factors 13 are compared with the pre-existing data set (such as the pre-existing data in the simplified example in Tables 1 to 3 in Figs. 9 to 11).

[0079] The method further includes determining 150 roadworthiness 19 of the bicycle based on the comparison. In one example, the comparison 17 includes the set of risk scores associated with each of the contextual factors 11 and damage factors 13. Thus determining 150 roadworthiness 19 includes a summation of the risk scores in the comparison. The sum of the risk scores (total risk score) may then be compared to risk ratings 73. As illustrated in Fig. 12, risk ratings may involve bands of total risk scores with an associated total risk that is indicative of roadworthiness 19.

[0080] The method 100 further includes displaying 160 an indication 27 of the roadworthiness 19 of the bicycle 1. This can include presenting the indication 27 of roadworthiness at a user interface 21 as illustrated in Fig. 7 which is a screenshot from a display. In some examples, this is generated in a report that can be sent to the technician, user of the bicycle, owner of the bicycle, manufacturer, insurance company, etc. This can include sending the report, comparison, and/or roadworthiness to the computer 53, 61, and/or server 63. [0081] Corrective action 25

[0082] In some examples, the method further includes determining corrective action 25 based on the comparison. This includes determining 151 potential corrective action 25 based on the comparison 17. The corrective action 25 can also be displayed 160 and form part of a report as noted above.

[0083] Determining 151 the potential corrective action 25 may involve comparison with the pre-existing data set. The pre-existing data set 15 may include association of the type, size or location of damage with the type of corrective action that can resolve such damage. In one simple example, this may include a data set where “crack damage” under “5mm” in size is associated with “repairable damage”. This can further include suggested repair type and/or method. Such specifics of the repair can include, specifying the size and type of carbon fibre fabric required for the repair, the amount of epoxy, repair technique, time, and other resources. In some examples, the suggested repair type and/or method may comprise a plurality of option for the suggested repair type/or method. In some examples, the method further includes an indication of respective resource requirements for those options. This enables the technician and/or user the choice to select from the options that suits their needs and situation.

[0084] It is to be appreciated that potential corrective action can take other forms. In some examples, this may include replacement 31 of one or more components of the bicycle. For example, if the damage is too severe, the pre-existing data may specify that the damage is non-repairable and the component must be replaced for the bicycle to be roadworthy.

[0085] In some examples, the potential corrective action includes information for a qualified repairer, supplier, specialist, manufacturer, etc. to facilitate repair, replacement, or servicing of components.

[0086] In another example, potential corrective action may include particular maintenance, or prescribing a maintenance schedule based on the contextual and damage factors. In further examples, potential corrective action may include additional or specific inspection, or inspection schedule for the bicycle as discussed in further detail below. [0087] Additional inspection

[0088] In some examples of the method, the result of the comparison includes additional inspection. For example, there may be a common causation (and/or correlation) between a damage type at a site that is related to potential damage at another site (that has not yet been detected). Thus in some examples, the method 100 includes determining 153 one or more areas of the bicycle for further inspection based on the comparison.

[0089] In one illustrative example, the pre-existing data set includes a relationship between a type of damage at the fork which increases the likelihood of corresponding additional damage at the steerer of the bicycle 1. In such cases, the method further includes determining 153 that if there is that type of damage at the fork, identifying those related areas that may have higher risk of damage and proposing additional inspections. It is to be appreciated that the related additional damage may be on the same component or other component of the bicycle.

[0090] To that end, the method includes displaying 161 an indication of the one or more areas for further inspection. This aids the system, and technician, to further scan 163 the one or more areas as a further inspection to detect additional potential damage. This can include, further scanning, review at a higher resolution, another scanning or inspection method, or different inspection schedule.

[0091] The method then includes categorising 165 the additional potential damage at the one or more areas 43 to generate an updated plurality of damage factors 13. The method further includes comparing the plurality of contextual factors 11 and the updated plurality of damage factors to form an updated comparison for determining roadworthiness 19, potential corrective action, and/or further inspection. F

[0092] Visualisation

[0093] In some examples, the method includes providing a visualisation to the technician, user at a graphical user interface. This may include an interactive report to illustrate damage at components of the bicycle. In some examples, this can include visualisation to assist identification of the locations of the damage, and guide further scanning, inspection, repair, and/or replacement. [0094] An example of an application that contains an interactive visualisation 77 at a graphical user interface is shown in Fig. 8. In this example, the visualisation includes a 3D graphical representation of the bicycle 1 with an overlay indicating locations of potential damage 3. When the technician wishes to scan a specific part of the bicycle (e.g. after they find some visible impact damage), they will be able to click the appropriate location on the 3D model, perform a scan on the real-world bicycle in that location, the data will be passed through to the application, and finally, a 3D visualisation will be created from that data and will be stored/tagged at that location. This may indicate areas of the bicycle that may need further corrective action. In addition, the user interface may be interactive to enable the technician or user to obtain further details, such as the results of the scan. In Fig. 8, this includes further data and information 79 relating to delamination damage that was detected at part of the bicycle.

[0095] In some examples of the application, the technician will be able to supplement this inspection data by selecting an area of the 3D bicycle model visualised within the application, and attaching/tagging the data obtained from the scanner or other instrument to this specific location.

[0096] The data obtained from contextual factors and scans (that may include detected damage and/or damage factors) could be used to produce risk visualisation(s).

[0097] Where large areas need to be scanned such as an entire downtube, the technician will be able to select the entire downtube, this will "unwrap" the image of the downtube into various segments (to conceptualise unwrapping, think about how a surface of a cube can be cut and laid out flat), the scan can be performed on each segment, and finally, the segments can be re-wrapped onto the original 3D model.

[0098] Further iterations of this could include some automated visualisation of stress on the bicycle when the data visualisation shows areas of concern. For example, a scan performed at the bottom of the down-tube that has areas of concern, might result in a visualisation of increased stress on closely connected areas such as around the bottom bracket and bottom of the seat tube. [0099] Initial model and pre-existing data set

[0100] To implement the model (as well us updates to the model using machine-learning), an initial pre-existing data set may be used. Data from initial inspections are obtained and provided to subject matter experts in non-destructive testing, bicycle mechanics/repair, and carbon fibre repair. These experts will:

- Assess the risk posed by each combination of contextual and damage factors observed in each inspection and provide a risk rating

- Assess the damage type(s) and suggest appropriate repair and/or maintenance options

- Assess the damage type(s) and suggest appropriate repair costs

[0101] Each of these represents an expert assessment/classification/label, which will be recorded to provide initial pre-existing data sets. In time, training data from each inspection (discussed below) may be used to develop supervised machine-learning models that predict the classification of each of these categories from contextual and damage factors recorded by technicians in future inspections.

[0102] Training

[0103] In some examples the pre-existing data set 15 is trained and may be continuously updated from learning patterns from received contextual factors, scans, and determined damage factors. This training can improve the model and performance of the inspection method to be more accurate and, potentially, predict weaknesses and recommend preventative measures. This includes improving the risk rating, the repair type, the repair cost, the potential risk posed if factors are modified, suggested maintenance schedule, and suggested follow-up inspection schedule.

[0104] In some examples, potential damage detected at the sites and/or the comparison (along with associated information such as contextual factors and other damage) is recorded and stored in a database 16. This may include storing the information at the computer or at the server 63. This can include an operator of the database and machine learning system. This forms historical records that may be used as training data to update the models. In some examples, the information is de-identified to preserve privacy of the users and owners of the bicycles.

[0105] In some examples, the pre-existing data set 15 is updated 170 based on comparing the pre-existing data set 15 with the received plurality of contextual factors 11 and plurality of damage factor 13. In further examples, this is achieved by comparing with a plurality of previous inspections and includes comparing the historical records (of the plurality of contextual factors and/or damage factors). This can be done, for example, at the computer and/or server. In some further examples, updating the pre-existing data set is continuously, whilst in other examples, this is done periodically.

[0106] Updating the pre-existing data set can be important to improve the model based on usage and other factors. For example, the original pre-existing data set may provide an inspection assessment with a particular roadworthiness 19 or prescribed corrective action 25. However, overtime there may be certain improvements (such as improvement in manufacture) than means the bicycle is more resilient or reliable even with such damage. Thus after a while (and with multiple inspections) it is evident that a particular type of damage/crack at a site does not actually affect overall risk over longer term use and the model will downgrade the risk score for that particular type of damage. In other examples, a particular model of bicycle (contextual factor) may be more durable than others, and the risk score may require adjustment.

[0107] Machine learning

[0108] Machine -learning includes a broad range of models, of which categories are supervised and unsupervised. Both supervised and unsupervised modelling approaches are described herein and may be used with the method and system.

[0109] First, supervised learning techniques such as regression, decision trees and random forests will be used to learn associations between inputs (contextual and damage factors recorded during inspections) and outputs (risk and repair classifications provided by subject matter experts). These models will then be used to produce predictions on future data. For example, when a technician inspects future bicycles and records various contextual and damage factors, the data will be processed by the ML algorithm(s), and the appropriate outputs/predictions will be automatically generated. Ultimately, this technical solution will enable technicians to produce high quality inspection reports informed by machine learning.

[0110] Second, supervised learning techniques will be used to predict the change in risk if specific factors were modified. For example, if the rider weight is reduced or the bicycle usage is changed to a different usage type, then an array of potential decisions can be provided to the client along with the potential changes in risk that would be expected if those changes were made.

[0111] Thus with supervised learning, updating the pre-existing data set 15 may further comprise sending, to a user display: at least part of the pre-existing data set; a representation, or summary, of the historical plurality of contextual factors and/or plurality of damage factors; and at least one or more proposed updates to the pre-existing data set.

[0112] This enables the subject matter expert to review the proposed updates and consider the historical data. In response, the subject matter expert can provide an acceptance, modification, or rejection to one or more of the proposed updates to the pre-existing data set 15.

[0113] Unsupervised learning models such as Apriori algorithm will be used to identify clusters of data and association rules, primarily from the inspection data of actual bicycle. That is, not being reliant on the classifications or input from the subject matter experts. These models will be used to learn patterns of factors that occur concurrently, so that maintenance and subsequent inspection schedules can be suggested. For example, unsupervised learning models can identify combinations of factors that frequently occur together (clusters) such as a specific damage type, damage size and damage location that frequently occurs in conjunction with a specific bicycle model and when the rider is above a certain weight (to clarify, perhaps there is a specific bicycle model that tends to exhibit fibre breakages near the bottom bracket when the bicycle is used for serious racing by riders above 100 kg). In future inspections where similar factors are recorded by a technician, the model can suggest a targeted strategy of subsequent inspections and maintenance to monitor for the damage type/size/location associated with these factors.

[0114] It is intended that the machine-learning models will be dynamic. Specifically, as more inspections are completed, the new data will be added to the pre-existing data set and database, and will be processed in the same manner as above to develop more accurate predictions over time.

[0115] Additionally, each bicycle can be tracked longitudinally through subsequent inspections (e.g. using the bicycle’s serial number) to monitor whether predictions made by the system were ultimately successful or unsuccessful, as well as whether any maintenance or repair solution was successful or unsuccessful. These outcomes can be implemented as aspects of reinforcement learning in future i.e. to penalise the model when it performs poorly and stimulate further improvements.

[0116] Additional data for the models

[0117] In addition, laboratory testing can help improve the modelling process with objective data. For example, a bicycle frame can have specific damage types applied at various points and then be tested to failure in a laboratory setting. We can test the difference in load tolerance between two identical frames, each with different damage types. This may help to refine the current model and ML aspects and the pre-existing data set.

[0118] If customer decisions are collected over time (e.g. did the customer replace or repair the frame or component as suggested), then a model of customer decision-making can also be developed. This may help to predict customer behaviours and propose better/altemative solutions. For example, if a customer is unlikely to replace a frame due to predicted cost, then perhaps alternative solutions can be provided, along with the expected risk of that alternative solution.

[0119] Variation of a computer application and user interface

[0120] Figs. 13 and 14 illustrate a variation of the computer application and user interface described above. In this example, potential damage and the plurality of sites 5 of the bicycle 1 can be selected or deselected using check boxes 81. As illustrated in Fig. 14, the sites “head tube”, “seat tube”, and “steerer” have been selected at the check boxes 81. Details of potential damage 3 for the selected sites 5 are shown at the user interface. In this example, the potential damage 3 is categorised in more detail compared to the example of Fig. 6 discussed earlier. This includes, for example the: part that is damaged, type of damage, location of the damage (e.g. side and/or area), the length of damage, the width of damage, and depth of damage. This also include an interface showing multiple damage on the same site 5 (e.g. both the head tube and seat tube have two separate potential damage).

[0121] Processing device

[0122] Fig. 15 illustrates an example of a processing device 55 than may be associated with a computer 53, server 63, and portable communication device 61. The processing device includes a processor 1510, a memory 1520 and an interface device 1540 that communicate with each other via a bus 1530. The memory 1520 stores instructions and data for implementing one or more of the methods 100 described above, and the processor 1510 performs the instructions (such as a computer program) from the memory 1520 to implement the methods 100. The interface device 1540 may include a communications module that facilitates communication with the communications network 59 and, in some examples, with the user interface and peripherals such as data store 16 or a display. It should be noted that although the processing device may be independent network elements, the processing device may also be part of another network element. Further, some functions performed by the processing device may be distributed between multiple network elements. For example, the server 63 may be associated with multiple processing devices and steps of the method 100 may be performed, and distributed, across more than one of these devices.

[0123] It will be appreciated by persons skilled in the art that numerous variations and/or modifications may be made to the above-described embodiments, without departing from the broad general scope of the present disclosure. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive.