FR3158809A1 - QUALITY CONTROL INSTALLATION AND PROCESS WITH ADAPTIVE INSPECTION SCHEME - Google Patents

Abstract

The invention relates to a quality control system and method intended to be implemented on a series of components resulting from an industrial process using adaptive inspection schemes. Figure to be published with the abstract: Figure 1

Description

QUALITY CONTROL INSTALLATION AND PROCESS WITH ADAPTIVE INSPECTION SCHEME

The invention lies in the field of quality control of industrial production processes and preferably relates to artificial intelligence technologies applied to said quality control.

Industrial manufacturing processes, for example in the automotive industry, are largely automated, but must nevertheless be supplemented by quality control processes for the operations carried out. This means that production defects are identified and isolated, and defective parts are repaired or removed. Each manufacturing step has its own quality control process, for which various solutions can be implemented, ranging from visual inspection by an operator to partial or complete automation.

For example, in the case of the painting stage in the automotive industry, the different types of vehicles produced in a factory converge on the painting line, where different colors are applied according to established patterns. On the same painting line, different body models can thus follow one another, each of which can be painted in a wide variety of colors, with a variable production frequency for each color.

To ensure that the painting process meets defined quality standards, a control plan is followed, including inspections of specific points. Since manual inspection is costly in terms of personnel and time, and also prone to human error, other known quality control systems can be implemented, involving the use of sensors and measuring systems to detect possible deviations in the manufacturing process.

Different types of data can be collected by different devices, such as paint thickness, and/or color and appearance measurements. Some inspection systems use data analysis techniques to monitor product quality and detect defects. An inspection scheme defines the control points to be inspected, with each point representing a sample.

However, the production line commonly sees a multitude of body colors and shapes, while automatic inspections, comprising several checks using different techniques, are carried out according to a fixed inspection pattern. Commonly used quality control techniques are therefore unable to adjust the inspection pattern in real time to suit each specific vehicle.

Thus, a control based on a set of fixed control points may lack adaptability to changes in environmental conditions or production processes and ultimately be ineffective in detecting defects.

Additionally, fixed control based on fixed frequency sampling may result in bias towards certain components or areas, which may cause a lack of representativeness of the collected data.

Such an inspection scheme can be costly in terms of resources and time, as some parameters are not necessarily relevant for all models to be inspected, which amounts to unnecessarily increasing the inspection time for certain vehicles in particular, and beyond that the inspection time for the entire production line.

Finally, control processes can be prone to human error in selecting control points or focusing on details, which can lead to incorrect or incomplete data collection.

Recently, techniques have emerged using machine learning, many of which are or can be applied to a quality control process in industrial production.

US20200166909A1 describes a method for controlling and optimizing a production process in real time using machine learning techniques. This method uses data collected by sensors to predict the quality of final products, and allows for adjusting production process parameters and providing rapid response to changes in variables in protocols.

CA3064593C also describes a method for controlling and optimizing manufacturing processes, with a particular focus on additive manufacturing. Machine learning models analyze input data and make accurate predictions in real time to adjust process parameters to optimize the quality of the final product.

US9294113B2 describes a technique for reducing power consumption in data acquisition systems using sensors or measuring devices. This technique uses a non-uniform adaptive sampling approach that acquires data only as needed, rather than at fixed times.

Document US11392837B2 focuses on the use of machine learning techniques for data analysis and data quality assessment in real time.

In summary, sampling solutions with a fixed inspection scheme can present problems of inefficiency, lack of adaptability, bias, costs and human errors. Methods implementing such solutions can therefore limit the ability to provide accurate and reliable information on the industrial process concerned, which can lead to poor decision-making and unnecessary costs for companies, while not reducing inspection time. A solution therefore remains to be found to overcome the problems and drawbacks encountered in the prior art.

One of the objectives of the invention is to propose a quality control method making it possible to adapt to the changing conditions of an industrial process and/or to reduce costs while minimizing human errors.

To this end, and according to a first aspect, the invention relates to a quality control method intended to be implemented on a series of entities resulting from an industrial process, the method being remarkable in that the series of entities comprises at least two entities and in that the method comprises the following steps:

select an entity for quality control

(b) determine an inspection scheme comprising a set of control points;

(c) measure, on the entity, one or more parameters indicating the quality of said process at the level of the set of control points defined by the inspection plan and collect the data, the measurement(s) being carried out by one or more measuring devices;

d) compare the data obtained in the previous step with at least one predefined reference interval for each of the measured parameters and at each of the control points, and generate a signal of conformity or anomaly or indication of a significant variation depending on the result of this comparison;

e) repeating the preceding steps a) to d) on one or more subsequent entities with the difference that, for at least one of the subsequent entities, step b) comprises the determination of a new inspection scheme comprising a new set of control points; the new inspection scheme being different from the previous one in that at least one control point present in the previous set is absent from the new set and/or in that at least one control point present in the new set is absent from the previous set; preferably, the determination of a new inspection scheme is done for each of the subsequent entities.

As will be understood from reading the definition just given, the invention proposes a remarkable quality control method in that the inspection schemes used successively are different, so as to adapt from one entity to another. This adaptation can be done in different ways as will be seen in detail.

According to a preferred implementation, step a) comprises a sub-step of identifying a type associated with the entity; preferably, said identification of the type associated with the entity is done by at least one means chosen from the reading of an identification code, recognition by a camera of the model of the entity using a pre-recorded directory of models, and/or a colorimetric test making it possible to define the color of the entity. By identifying a type associated with the entity, it is possible to verify the quality of different types at the level of the same installation while adapting the verification scheme to said type of entity. In the context of an industrial painting process, a type of entity may comprise a given color of the entity and/or a given model of the entity.

Preferably, step a) comprises a sub-step of identifying a type associated with the entity and said selection is made according to a selection frequency specific to the type of entity; said identification of the type associated with the entity being made by at least one means chosen from reading an identification code, recognition by a camera of the model of the entity using a pre-recorded directory of models, and/or a colorimetric test making it possible to define the color of the entity. It is thus possible to select certain types of entities more frequently than others. A higher selection frequency can thus be used for minority types or for types most likely to show anomalies.

According to a preferred implementation, step b) comprises providing a catalog defining a total number of control points, and the set of control points constitutes a selection of a given number of control points in said catalog such that the number of control points in the set is less than or equal to the total number of control points defined in the catalog. The invention makes it possible to no longer carry out measurements at the level of all existing control points but at the level of a selection of the latter, which makes the verification of the quality of an entity faster.

According to a preferred implementation, step b) of determining an inspection scheme is carried out by means of a machine learning algorithm or by selecting an inspection scheme from a collection of pre-recorded inspection schemes. The use of a machine learning algorithm is advantageous due to the possibilities of adapting the inspection scheme that it generates.

According to a preferred implementation, step a) comprises a sub-step of identifying a type associated with the entity, and step b) comprises providing a catalog defining a total number of control points and a frequency of appearance of each control point of the catalog in an inspection scheme; and when the selected entity is of the same type as one or more previously selected entities, the definition of the new inspection scheme takes into account said frequency of appearance in order to add and/or remove control points compared to the previous inspection scheme; preferably, step b) comprises a sub-step of updating the frequency of appearance associated with each of the control points. It is understood that all of the control points will be checked not on one entity but on a plurality of entities. By reducing the inspection time of the entities, it is possible to check more entities in a given time.

Advantageously, when initiating the process, step b) comprises:

the provision of a catalog defining a total number of control points;

the provision of an initial set of training data comprising simulation data of the industrial process and/or historical data obtained by a qualified operator or by a pre-existing control process on said industrial process;

defining an optimal set of control points defining an initial inspection scheme by implementing a machine learning algorithm that has been trained using the initial training data set; and

the choice of the initial inspection scheme as the inspection scheme;

preferably, step a) comprises a sub-step of identifying a type associated with the entity and the initial set of training data provided in step b) is specific to the type of said entity.

The start-up of the process requires the determination of an initial inspection scheme, it is advantageous that this initial scheme is optimal, this can be achieved by means of artificial intelligence.

According to a preferred implementation, when the comparison carried out in step d) generates an anomaly signal or an indication of a significant variation in a parameter at a given control point, the following inspection scheme will include said control point; preferably, step a) includes a sub-step of identifying a type associated with the entity and the following inspection scheme including said control point is an inspection scheme for an entity of the same type. The method thus makes it possible to monitor the evolution of significant variations in the data collected (or the appearance of an anomaly or a defect) on a series of entities. The invention therefore makes it possible to intervene as quickly as possible before the appearance of anomalies or defects and/or to verify that measures have been taken upstream to correct the anomalies.

According to a preferred implementation, the method comprises a preliminary step of creating a digital twin for each of said entities so as to obtain digital twin simulation data for at least one of the parameters that can be measured on said entities, and step a) comprises the sub-steps of identifying the entity; comparing the digital twin simulation data to predefined reference intervals; and selecting or not selecting said entity based on the result of the comparison. Preferably, when the entity is selected, step b) comprises identifying one or more control points revealing an anomaly or a significant variation of a parameter by comparing the digital twin simulation data to predefined reference intervals, and inserting into the inspection scheme determined for said entity the control point(s) thus identified. The use of digital twins facilitates the identification of the type of entity and/or an intelligent selection of entities showing or likely to show anomalies and/or significant variations on certain parameters.

Advantageously, the industrial process is a painting process and one or more of the parameters indicating the quality of said process are chosen from the thickness of the paint layer, the colorimetry, the gloss, the presence of appearance defects, the presence of tin drips and the presence of appearance defects resulting from a stamping defect.

According to a second aspect, the invention relates to a computer-readable medium comprising computer-executable code for implementing a method according to the first aspect.

According to a third aspect, the invention relates to an installation for implementing a quality control method for an industrial process comprising an inspection station comprising a robot carrying, in its hand, an inspection head comprising at least one measuring tool adapted to measuring one or more parameters indicating the quality of said process, the installation being remarkable in that it further comprises a system for controlling a method according to the first aspect comprising a data acquisition unit and at least one computer.

Preferably, the industrial process is a painting process and the inspection head comprises at least one measuring tool selected from a sensor for measuring the thickness of the paint layer, a colorimetric sensor, a vision system for controlling the appearance preferably associated with a lighting system. For example, the lighting system comprises an ultraviolet lamp.

For example, the sensor for measuring the thickness of the paint layer is a terahertz spectroscopy probe, preferably a pulsed terahertz imaging probe.

For example, the colorimetric sensor combines the functions of a color densitometer and a color photometer

For example, the vision system comprises a laser profilometer and one or more cameras; preferably, at least one camera may be a stereoscopic camera allowing the restitution of a relief image.

The invention will be well understood and other aspects and advantages will appear clearly on reading the following description, given by way of example with reference to the attached drawing board on which:

There FIG. 1 represents an installation according to the invention.

In the following description, the term “comprise” is synonymous with “include” and is not limiting in that it allows the presence of other elements in the installation or process to which it relates. It is understood that the term “comprise” includes the terms “consist of”. In the different figures, the same references designate identical or similar elements.

The invention relates to a quality control method and the installation implementing such a method.

The invention relates to a quality control method intended to be implemented on a series of entities resulting from an industrial process, the method being remarkable in that the series of entities comprises at least two entities and in that the method comprises the following steps:

(a) select an entity for quality control

(b) determine an inspection scheme comprising a set of control points;

(c) measure, on the entity, one or more parameters indicating the quality of said process at the level of the set of control points defined by the inspection plan and collect the data, the measurement(s) being carried out by one or more measuring devices;

d) compare the data obtained in the previous step with at least one predefined reference interval for each of the measured parameters and at each of the control points, and generate a signal of conformity or anomaly or indication of a significant variation depending on the result of this comparison;

e) repeating the preceding steps a) to d) on one or more subsequent entities with the difference that, for at least one of the subsequent entities, step b) comprises the determination of a new inspection scheme comprising a new set of control points; the new inspection scheme being different from the previous one in that at least one control point present in the previous set is absent from the new set and/or in that at least one control point present in the new set is absent from the previous set; preferably, the determination of a new inspection scheme is done for each of the subsequent entities.

The method according to the invention can be used for the quality control of various industrial processes. Preferably, it is used in the context of the quality control of an industrial process for painting a vehicle body or a sub-assembly of a vehicle body.

One or more of the parameters indicating the quality of said process may be chosen according to the industrial process concerned. In the context of a painting process, one or more of the parameters measured in step b) are chosen from the thickness of the paint layer, the colorimetry, the gloss, the presence of appearance defects, the presence of tin drips and the presence of appearance defects resulting from a stamping defect.

As will be seen in detail, the invention proposes an intelligent method which makes it possible to adapt the inspection scheme associated with the quality control of an industrial process to the changing conditions of said industrial process and/or which makes it possible to test certain control points in an alternative or sequential manner so as to reduce the inspection time.

Preferably, step a) includes a sub-step of identifying a type associated with the entity.

Advantageously, step b) comprises providing a catalog defining a total number of control points. This catalog of control points can be established by relying on historical data to identify the areas that represent critical points in the production process. The historical data provides information on the critical areas in which problems have occurred in the past and in which they are likely to occur in the future. On the other hand, critical areas are identified by assessing the risk associated with each stage of the production process. When historical data is not available, or in addition to existing historical data, the catalog of control points can be established from simulations of said industrial process. Indeed, careful identification of control points is essential to ensure accurate and reliable quality control results.

The control point catalog lists all the control points that may be relevant in the context of setting up the quality control process. In one embodiment, the set of control points of the initial inspection scheme (i.e., the first set used when starting the process) includes all the control points in the catalog. But preferably, the initial set of control points and the following sets constitute selections of certain control points in the catalog.

Thus, according to a preferred embodiment, step b) comprises providing a catalog defining a total number of control points, and the set of control points constitutes a selection of a given number of control points in said catalog such that the number of control points in the set is less than or equal to the total number of control points defined in the catalog. It will be understood that the set of control points may be that of an initial inspection scheme (i.e. initiating the implementation of the method) or an inspection scheme adapted from an inspection scheme previously used in the context of the method.

When the inspection scheme is the initial inspection scheme, the selection of the control points forming the initial set of control points can be done by the instruction given by an operator by means of a human-machine interface, or by the implementation of a machine learning algorithm (i.e. an artificial intelligence algorithm) by means of an initial learning data set comprising simulation data of the industrial process; and/or historical data obtained by a qualified operator or by a pre-existing control method on said industrial process. According to a preferred embodiment of the invention in which different types of entities are checked within the same installation, it will be advantageous for the initial set of learning data provided to be specific to the type of the entity.

Thus, according to a preferred embodiment, step b) comprises:

the provision of a catalog defining a total number of control points;

the provision of an initial set of training data comprising simulation data of the industrial process, and/or historical data obtained by a qualified operator or by a pre-existing control process on said industrial process;

defining an optimal set of control points defining an initial inspection scheme by implementing a machine learning algorithm that has been trained using the initial training data set; and

the provision of the initial inspection scheme as the inspection scheme.

Preferably, step a) comprises a sub-step of identifying a type associated with the entity and the initial set of training data provided in step b) is specific to the type of said entity.

It will be understood that once the process has been initiated, it will allow the inspection scheme used in the quality control process to be adapted regularly (for example according to a given frequency), and preferably continuously or in real time. A new set of learning data is obtained from the data collected in measurement step c) and replaces or adds to the initial set of learning data.

This adaptation of the inspection plan can be done at several levels.

Adaptation of the inspection scheme to entities according to their type

The adaptation can be done at step b) of determining an inspection scheme. Indeed, it will be advantageous for a person skilled in the art, when adapting the inspection scheme, for the method to take into account factors specific to the industrial process concerned. In the case of an industrial process for painting a vehicle body or a sub-assembly of a vehicle body, the factors specific to said industrial process may include the model of vehicle concerned and/or its color.

Therefore, when the industrial process is a painting process implemented on one or more vehicle body models and/or vehicle body subassemblies, it is possible to associate an inspection scheme with a given model and/or the color applied to the model(s). The step of providing an inspection scheme therefore comprises the selection of an inspection scheme (i.e. specific and/or relevant control points) for the type of entity whose quality is controlled, said type being able to be determined by its model and/or its color.

According to a preferred embodiment, step a) comprises the sub-step of identifying the type of the entity whose quality is verified, and step b) comprises determining an inspection scheme specific to the type of said entity; preferably, said identification is done by recognizing the model of the entity from a pre-recorded directory of models and/or by a colorimetric test making it possible to define the color of the entity. The selection of an inspection scheme associated with said entity can be done from a collection of pre-recorded inspection schemes or can constitute the determination of said inspection scheme by a machine learning algorithm.

Recognition of the entity model can be done by any known means, for example by visual recognition means or by reading an identification code carried by the entity. The colorimetric test can be done by any known means and can also be replaced by reading an identification code carried by the entity. The identification code can be a barcode or a QR code applied to the entity with invisible ink.

Thus, the process makes it possible to apply an inspection scheme that is adapted to the entity whose quality is verified by identifying the type of said entity. It is therefore possible to test different types of entities on the same inspection station while carrying out a specific inspection adapted to each of the entities.

Adaptation of the selection frequency of entities according to their type

Identifying entities by type offers another advantage. Often, not all entities produced as part of the industrial process are inspected. Inspection is carried out on only certain entities, chosen to be representative of the said process. When the quality control process is applied to only some of the entities produced, it will be advantageous for the entity identification sub-step to be carried out on all the entities produced and not only on the verified entities. This makes it possible to select the entities to be verified in such a way as to favor the selection of the least frequent types of entities. The process can therefore also define a frequency of selection (i.e. inspection) of entities according to their type.

For example, step a) comprises a sub-step of identifying a type associated with the entity and said selection is made according to a selection frequency specific to the type of entity; said identification of the type associated with the entity being made by at least one means chosen from

reading an identification code;

recognition by a camera of the entity model using a pre-recorded model directory; and/or

a colorimetric test to define the color of the entity.

The selection can therefore be made according to a predefined selection frequency associated with its type, however, the frequency of selection of a type of entity for its control can be increased or decreased according to predefined factors.

Indeed, and as is well known, some vehicle colors are more represented than others. For example, more white cars (the majority color) are produced than green cars (the minority color). It may therefore be interesting to increase the selection frequency of entities showing a minority color in order to increase the data collection associated with said color. The same goes for vehicle models.

Thus, the process makes it possible to apply a frequency of selection of entities according to their type to submit them to quality control.

Adaptation of the inspection time of an entity

The method makes it possible to adapt or reduce the inspection time per entity. To this end, and taking advantage of the fact that the inspection schemes used include a number of control points lower than the total number of control points listed in the catalog, the method makes it possible to successively use different inspection schemes on entities of the same type. Consequently, all the control points will be checked not on a single entity but on a plurality of entities. For this, each control point can be associated with a given verification frequency (or frequency of appearance in an inspection scheme), the frequency being able to be weighted manually or by the data collected on the previous entities by the machine learning algorithm. In particular, when certain control points are identified as giving generally satisfactory results, their frequency of appearance in an inspection scheme can be reduced.

Thus, for example, if the catalogue includes 30 control points, it is possible to limit the inspection schemes to 15 control points (or to a variable number chosen between 10 and 20) and to assign a frequency of appearance to said control points so that all the control points are verified on a given number of entities successively subject to quality control (for example 2, 3 or 4).

It is understood that with a reduced number of control points, the inspection time of an entity is reduced. The quality of the entities tested at a given control point is statistical. This implementation is also interesting when the quality control process checks not all the entities produced but a part of them, since with a reduced inspection time it is possible to check more entities. A compromise is found between increasing the number of verified entities and reducing the inspection time.

Thus, and according to a preferred embodiment, step b) comprises providing a catalog defining a total number of control points and a frequency of appearance of each control point of the catalog in an inspection scheme; and when the selected entity is of the same type as one or more previously selected entities, the definition of the new inspection scheme takes into account said frequency of appearance in order to add and/or remove control points compared to the previous inspection scheme. Preferably, step b) comprises a sub-step of updating the frequency of appearance associated with each of the control points. Advantageously, the frequency of appearance of each control point is defined by the machine learning algorithm using the initial set of training data initially and, subsequently, using the new set of training data comprising the data collected in step c).

Preferably, step b) includes a sub-step of updating the frequency of occurrence associated with each of the control points based on the data collected on the previous entities. When a control point is associated with the maximum frequency of occurrence, it will be present in all the inspection schemes used. A maximum frequency of occurrence can be defined for control points in critical areas.

For other control points, when the collected data consistently show satisfactory results, it is possible to reduce their frequency of occurrence in an inspection scheme. However, it would be advantageous to define a minimum frequency of occurrence for a control point in an inspection scheme so that it is regularly present in the inspection schemes used. The frequency of occurrence of a given control point can be increased again when significant variations or defects (or anomalies) are detected in the measurements carried out on the previous entity(ies).

Adaptation to the results of the measurements taken

The method allows for continuous (i.e., real-time) or regular adaptation of the inspection scheme for each given type of entity based on variations in the industrial process whose quality is being verified and environmental conditions. Thus, the method can use the machine learning algorithm to identify, in relation to the data collected during the application of the previous inspection scheme, which control points are relevant in the new inspection scheme (or subsequent inspection scheme).

Thus, for example, step d) includes comparing the result of the measurements carried out on the parameter(s) tested during step c) to at least one predefined reference interval for each of the parameters, and/or to the data from a digital twin simulation. When the result of a measurement is outside the predefined interval(s), a defect (i.e. an anomaly) is detected. It is understood that the defect is then reported. However, advantageously, the detection of a defect will also result in the repeating of the presence of the control point that made it possible to identify the defect in the following inspection scheme in order to be able to determine whether the defect is isolated or recurring. A recurring defect in fact requires an adaptation of the industrial process upstream. The repeating of the presence of a control point in the following inspection scheme is then done regardless of its frequency of occurrence or by making this frequency of occurrence maximum.

Furthermore, step d) may include comparing the result of the measurements carried out on the parameter(s) tested during step c) with one or more of the results previously obtained for the same parameter(s) at the same control point, and/or with the data from a digital twin simulation. When a significant variation is detected, the method causes the presence of the control point that made it possible to identify said significant variation to be repeated in the following inspection scheme (regardless of its frequency of occurrence) in order to be able to monitor the evolution of the variation and, if this significant variation persists or increases, to report this variation before the appearance of a defect (i.e. an anomaly). To identify a variation as being significant, it is possible to define at least two reference intervals for the measurement of the same parameter, the two intervals being nested within each other and being centered on the same median value. When the result of a measurement is included in both intervals, the variation is said to be insignificant. When the result of a measurement is included in only one of the two intervals (i.e., in the wider of the two), the variation is said to be significant.

Thus, preferably, step d) comprises comparing the result of the measurements carried out on the parameter(s) tested during step c) to at least one predefined reference interval for each of the parameters, and/or to the data of a digital twin simulation; identifying, where appropriate, one or more control points revealing a defect or a significant variation of a parameter, and step b) implemented for the next entity comprises inserting into the new inspection scheme the control point(s) thus identified. The next entity considered may or may not be of the same type; preferably, it is of the same type.

Furthermore, the detection of a defect or significant (i.e., large) variations at a given control point may result in the insertion of one or more associated additional control points into the new inspection scheme, said associated additional control points showing a significant probability of the presence of a defect or significant variation when a defect or significant variation is detected at said given control point. The association relationship may be predetermined using historical or simulation data, or may be determined by the machine learning algorithm by using the initial training data set first, and subsequently by using the new training data set. The invention therefore provides a predictive tool on the occurrence of defects by studying the measured parameters and their variations.

It is understood that the implementation of the method according to the invention, with an adaptive inspection scheme, allows both a reinforced inspection at the level of the zones and/or parameters identified as sensitive, and at the same time an inspection reduced to the bare minimum in certain zones and/or at the level of the parameters identified as satisfactory. The implementation of an adaptive inspection scheme therefore makes it possible to reduce the inspection time of an entity while refining said inspection and making it possible to detect as early as possible the defects generated by the variations of the industrial process.

Adaptation using a digital twin

According to a preferred embodiment, the method further comprises a preliminary step of creating a digital twin for each of said entities so as to obtain digital twin simulation data for at least one of the parameters that can be measured on said entities.

Advantageously, the method can also implement a digital twin. The creation of a digital twin (in English "digital twin" or "device shadow") is known. It is a digital model that faithfully reconstructs an entity (in the form of a virtual clone). The digital twin is not only a perfect replica of the entity at the time it arrives at the inspection station but it also includes information on the industrial process(es) it has passed through since its design. Also, the determination of an optimal inspection scheme for an entity can take into account the information provided by the digital twin. Indeed, the data from the digital twin simulation can be compared to the predefined reference data and, in the event of a variation in one or more parameters, the control points corresponding to said parameters can then be selected to verify the reality of the variation or not.

Alternatively or additionally, the method according to the invention will compare the data from a digital twin simulation with the actual measurements of the parameters obtained in step c) and/or with reference data.

Indeed, when using a digital twin, the person skilled in the art will benefit from ensuring that the simulation data of said digital twin are:

compared to predefined reference intervals, in a predictive framework for detecting defects and significant variations and therefore adapting the inspection scheme; and/or

compared to measurements made on the entity in question to provide training data to refine the digital twin simulation.

Thus, the invention comprises creating a digital twin for said entities so as to obtain digital twin simulation data for at least one of the parameters that can be measured on said entities, and step a) comprises the sub-steps of identifying the entity; comparing the digital twin simulation data to predefined reference intervals; and selecting or not selecting said entity for verification based on the result of the comparison. Thus, the selection of an entity likely to show a defect or a significant variation in a parameter can be carried out without taking into account the application of a selection frequency of said entity.

Preferably, when the entity is selected, the method comprises identifying one or more control points revealing a defect or a significant variation of a parameter by comparing the digital twin simulation data to predefined reference intervals, and inserting into the inspection scheme used for said entity the control point(s) thus identified. Thus, the inspection scheme is adapted to the simulation data obtained by the use of a digital twin and targets in particular the control points in which a problem is likely to exist.

The industrial process is preferably a painting process, and one or more of the parameters indicating the quality of said process are chosen from the thickness of the paint layer, the colorimetry, the gloss, the presence of appearance defects, the presence of tin drips and the presence of appearance defects resulting from a stamping defect.

The invention also relates to a computer-readable medium comprising computer-executable code for implementing a method as defined above.

Finally, the invention relates to an installation for implementing a quality control method for an industrial process comprising an inspection station comprising a robot carrying, in its hand, an inspection head comprising at least one measuring tool suitable for measuring one or more parameters indicating the quality of said process, the installation being remarkable in that it further comprises a system for controlling a process as described above.

There FIG. 1 illustrates an exemplary embodiment of an installation 1 according to the invention. The installation comprises an inspection station 3 comprising a robot 7 carrying, in its hand, an inspection head 9 comprising at least one measuring tool suitable for measuring one or more parameters indicating the quality of said process. The robot 7 is preferably poly-articulated. The robot 7 can be placed on a mobile gantry itself placed above the production line bringing the entities to be inspected, such as for example a vehicle body or a vehicle body subassembly. On the FIG. 1 , the robot 7 is arranged at the edge of the line. The installation 1 advantageously comprises a control system 11 comprising a unit for acquiring the data obtained by the robot 7 during an inspection cycle, and at least one computer. The data acquisition unit collects, stores and digitizes the data from the measuring tools carried by the robot 7. The data acquisition unit sends the data to at least one computer, which will integrate the data received via the data acquisition unit with one or more data sets chosen from one or more historical data sets, one or more reference data sets (i.e. one or more reference intervals and identification codes) and one or more data sets from the digital twin simulation. The computer has received executable code for implementing the method via a suitable readable medium. The computer is also programmed to control the robot 7.

The computer also makes it possible to operate at least one machine learning algorithm which uses data chosen from historical, reference, digital twin, and/or measurement data at control points; in order to define the inspection schemes and, preferably, to select the entities to be checked according to the method described above.

According to a preferred implementation of the invention, the computer is connected to an alert system, which notifies an operator in the event of detection of an anomaly or significant variation. According to a preferred implementation of the invention, a user interface allows the operator to access the data in real time and to carry out corrective actions either at the level of the installation implementing the industrial process upstream or at the level of a repair unit downstream.

The installation 1 is described in the context of the implementation of a quality control method for a painting process but may be adapted to other industrial processes. The robot may be equipped with a 5, 6 or 7-axis polyarticulated arm so as to show sufficient flexibility to be able to align the inspection head with all the control points to be inspected on the body 5 or the body subassembly of the vehicle. Preferably, the robot 7 has a 6-axis polyarticulated arm. The inspection head carries a series of measuring tools specific to each type of control. The measuring tools include at least one sensor for measuring the thickness of the paint layer, a colorimetric sensor for color control, and a vision system for appearance control, preferably associated with a lighting system.

The sensor for measuring the thickness of the paint layer may be, for example, an ultrasonic probe, an eddy current probe, an X-ray tomography probe, or another probe allowing non-destructive testing. Preferably, the sensor for measuring the thickness of the paint layer is a terahertz spectroscopy probe; preferably a pulsed terahertz imaging probe in the range of 0.1 to 10 THz.

The colorimetric sensor may be a color densitometer, or a color tristimulus photometer, or a spectrophotometer. Preferably, the colorimetric sensor will combine the functions of a color densitometer and a color photometer.

The vision system may include a laser profilometer and one or more cameras; at least one camera may be a stereoscopic camera allowing the restitution of a relief image.

Advantageously, the paint layer thickness measuring sensor, the colorimetric sensor and the vision system are oriented at different angles so that they can be used alternately after rotation of the inspection head.

According to the invention, prior to the steps of measuring parameters within the framework of the implementation of the quality control method, the robot 7 can carry out an identification of the entity by means of a determination of its color, its model, or by reading a verification code by a camera.

The invention is remarkable in that a control point as defined by the inspection scheme can be the subject of one or more types of measurements. For example, a control point can be the subject of a measurement of the thickness of the paint layer, and/or a colorimetry measurement, and/or an appearance measurement. The appropriate sensor(s) will be successively used on the given control point, as defined in the inspection scheme.

The robot 7 will position its inspection head 9 near the control point to be inspected. The inspection head 9 will be positioned in such a way that the appropriate sensor is in contact, or at a sufficient distance to be activated, with the control point.

For example, the inspection head 9 will be positioned so that the paint layer thickness measuring sensor is in contact, or at a sufficient distance to be activated, with the control point. After activation of said sensor, the paint layer thickness measurement will be taken and sent to the data acquisition unit.

Techniques for measuring the thickness of a paint layer are well known to those skilled in the art. The technique used in the context of the invention uses a terahertz imaging probe. Terahertz non-destructive testing technologies are capable of carrying out measurements by non-contact inspection, with high precision and good penetration. In particular, terahertz time domain spectroscopy (THz-TDS) technology has been described as a technique for non-destructive testing of thin layers of paint on automobiles; from this derives pulsed terahertz imaging (THz Pulse Imaging) in the range of 0.1 to 10 THz, for which sensors are commercially available. Other terahertz technologies in development are however conceivable, such as frequency domain spectroscopy. Preferably, the invention will use a pulsed terahertz imaging probe.

Preferably, the sensor for measuring the thickness of the paint layer is a terahertz imaging probe; preferably a pulsed terahertz imaging probe in the range of 0.1 to 10 THz.

If defined in the inspection scheme by the intelligent algorithm, the same inspection point can be subject to colorimetric analysis. The robot 7 will then rotate the inspection head 9 so as to place the colorimetric sensor at an appropriate distance. The colorimetric sensor will be activated and the corresponding measurement will be sent to the data acquisition unit.

Colorimetric analysis aims to detect the color of the entity, but can also ensure that the applied paint is the exact desired shade, or at least falls within defined spectral limits. A colorimeter determines the color based on the red, blue, and green components of the light absorbed by the sample. Sensitive to light, it measures the amount of color absorbed by an object or substance, as well as the transmittance for a small number of predetermined waves. The colorimetric sensor can be, for example, a color densitometer, or a color tristimulus photometer, or a spectrophotometer. Preferably, the colorimetric sensor will combine the functions of a color densitometer and a color photometer.

If defined in the inspection diagram, the same control point can be subject to an appearance check. The robot 7 will then rotate the inspection head so as to place the vision system, associated with the lighting system, at an appropriate distance. The vision system will be activated and the corresponding image(s) will be sent to the data acquisition unit.

The purpose of appearance control is to visualize possible defects in the application of the paint layer. The vision system may include a laser profilometer and one or more cameras; the camera may be a stereoscopic camera allowing the reproduction of a relief image.

Once the camera is aligned with the control point, an acquisition of one or more images of the control point is carried out. Preferably, the method comprises the acquisition of at least two images from different angles so as to be able to obtain a relief image of said control point. The images can be acquired successively with a modification of the angle and/or the positioning of the camera between two acquisitions. According to a preferred embodiment, the camera is a stereoscopic camera (or “3D camera” or “3-dimensional camera”) which makes it possible to carry out a double image acquisition simultaneously from two different angles or positions. Such cameras are known and generally comprise two lenses - placed side by side in a single housing. The camera will simultaneously acquire a pair of stereoscopic images, i.e. two twin (but not similar) images for the purpose of reproducing the relief (i.e. for the purpose of a three-dimensional inspection).

The lighting system, located on the inspection head so as to illuminate the control point to be photographed, is used to provide at least one standardized light of constant spectrum; possibly, several lights of different spectra can be used and give rise to several sequences of shots.

Laser profilometers are known to those skilled in the art, and use a laser beam to scan the surface of an object. By measuring the time it takes for the laser to return after hitting the surface, the device can calculate the distance and create a three-dimensional profile of the object.

As seen above, a camera can also be used in the context of reading an identification code using invisible ink and the lighting system can therefore include ultraviolet radiation lighting means making said identification code appear for reading.

Claims (10)

Quality control method intended to be implemented on a series of entities resulting from an industrial process, the method being characterized in that the series of entities comprises at least two entities and in that the method comprises the following steps:

• (a) select an entity for quality control
• (b) determine an inspection scheme comprising a set of control points;
• (c) measuring, on the entity, one or more parameters indicating the quality of said process at the level of the set of control points defined by the inspection plan and collecting the data, the measurement(s) being carried out by one or more measuring devices;
• (d) compare the data obtained in the previous step with at least one predefined reference interval for each of the measured parameters and at each of the control points, and generate a signal of conformity or anomaly or indication of a significant variation depending on the result of this comparison;
• e) repeating the preceding steps a) to d) on one or more subsequent entities with the difference that, for at least one of the subsequent entities, step b) comprises the determination of a new inspection scheme comprising a new set of control points; the new inspection scheme being different from the previous one in that at least one control point present in the previous set is absent from the new set and/or in that at least one control point present in the new set is absent from the previous set; preferably, the determination of a new inspection scheme is done for each of the subsequent entities.

Method according to claim 1, characterized in that step a) comprises a sub-step of identifying a type associated with the entity; and/or in that step a) comprises a sub-step of identifying a type associated with the entity and said selection is made according to a selection frequency specific to the type of entity; said identification of the type associated with the entity being made by at least one means chosen from

• reading an identification code;
• recognition by a camera of the entity model using a pre-recorded model directory; and/or
• a colorimetric test to define the color of the entity.

Method according to claim 1 or 2, characterized in that step b) comprises providing a catalog defining a total number of control points, and the set of control points constitutes a selection of a given number of control points in said catalog such that the number of control points in the set is less than or equal to the total number of control points defined in the catalog; and/or in that step b) of determining an inspection pattern is done by means of a machine learning algorithm or by selecting an inspection pattern from a pre-recorded collection of inspection patterns. Method according to claim 1 to 3, characterized in that step a) comprises a sub-step of identifying a type associated with the entity and step b) comprises providing a catalog defining a total number of control points and a frequency of appearance of each control point of the catalog in an inspection scheme; and in that when the selected entity is of the same type as one or more previously selected entities, the definition of the new inspection scheme takes into account said frequency of appearance in order to add and/or remove control points compared to the previous inspection scheme; preferably, step b) comprises a sub-step of updating the frequency of appearance associated with each of the control points. Method according to one of claims 1 to 4, characterized in that during the initiation of the method, step b) comprises:

• the provision of a catalog defining a total number of control points;
• the provision of an initial set of training data comprising simulation data of the industrial process, and/or historical data obtained by a qualified operator or by a pre-existing control process on said industrial process;
• defining an optimal set of checkpoints defining an initial inspection scheme by implementing a machine learning algorithm that has been trained using the initial training data set; and
• the choice of the initial inspection scheme as the inspection scheme;

preferably, step a) comprises a sub-step of identifying a type associated with the entity and the initial set of training data provided in step b) is specific to the type of the entity. Method according to one of claims 1 to 5, characterized in that the comparison carried out in step d) generates an anomaly signal or an indication of a significant variation of a parameter at a given control point, the following inspection scheme will include said control point; preferably, step a) includes a sub-step of identifying a type associated with the entity and the following inspection scheme concerned is an inspection scheme for an entity of the same type. Method according to one of claims 1 to 6, characterized in that it comprises a preliminary step of creating a digital twin for each of said entities so as to obtain digital twin simulation data for at least one of the parameters that can be measured on said entities, and in that step a) comprises the sub-steps of identifying the entity; comparing the digital twin simulation data to predefined reference intervals; and selecting or not selecting said entity depending on the result of the comparison; preferably, when the entity is selected, step b) comprises identifying one or more control points revealing an anomaly or a significant variation of a parameter by comparing the digital twin simulation data to predefined reference intervals, and inserting into the inspection scheme determined for said entity the control point(s) thus identified. Method according to one of claims 1 to 7, characterized in that the industrial process is a painting process and in that one or more of the parameters indicating the quality of said process are chosen from the thickness of the paint layer, the colorimetry, the gloss, the presence of appearance defects, the presence of tin drips and the presence of appearance defects resulting from a stamping defect. A computer-readable medium comprising computer-executable code for implementing a method according to one of claims 1 to 8. Installation (1) for implementing a quality control method for an industrial process comprising an inspection station (3) comprising a robot (7) carrying in its hand an inspection head (9) comprising at least one measuring tool suitable for measuring one or more parameters indicating the quality of said process, the installation (1) being characterized in that it further comprises a control system (11) for a method according to one of claims 1 to 8 comprising a data acquisition unit and at least one computer.