Method and system for quality control of an object in an apparatus for producing an object in a continuous cycle

Abstract

A method for quality control of objects in an apparatus for producing objects in a continuous cycle comprises the following steps: capturing an image (I) for each object (O); applying a first processing step (102A) to each image, which assigns the image and the corresponding object to one of two categories, namely a defective object category and a non-defective object category; if the image belongs to the defective object category, applying a second processing step (102B) to the image data associated with the image, and further classifying the image and the corresponding object according to a plurality of defect categories.

Description

Technical Field

[0001] The present invention relates to a method and a system for quality control of an object in an apparatus for producing an object in a continuous cycle.

Background Art

[0002] In a production line, especially in a high-output production line, quality control is of utmost importance.

[0003] In such a line, defective objects may occur, and a quality check on the objects must be performed before the objects are shipped from the factory so that the defective objects can be removed.

[0004] Quality control can consist of a manual visual inspection. However, since this method is not accurate enough, it is usually replaced by an automated visual inspection.

[0005] As prior art, methods for automatically detecting defects of an object are known. In these methods, one or more images of the object to be inspected are captured by an optical device, and based on the image data, the defects are specified as much as possible. Further, after the first process in which the defects are specified, these methods often also involve further data processing and analysis to classify the specified defects.

[0006] In this regard, Patent Documents US20090324057A1, CN110349150, CN110838107, US2013129185 and WO2004111618 describe automatic inspection methods for detecting defects. Patent Document US2021 / 010953A1 discloses a system for high-speed inspection and examination of an object using X-rays. This system focuses on the inspection of integrated circuits by analyzing various parts of the integrated circuit. However, this system is quite complex and does not allow for real-time (online) management of the quality of objects manufactured at a high production rate (this is a typical situation in the field of rigid packaging).

[0007] In fact, in this field, there is an increasing need for methods that can perform the quality control of objects more accurately and in a shorter time.

[0008] It must be said that the present invention can be applied to all fields where quality control of objects is required, for example, the field of rigid packaging. In this field, the products to be defect-checked may be made of plastic (caps, parisons, containers), or other materials (glass, aluminum, bottles, cans).

Summary of the Invention

[0009] The present disclosure aims to overcome the above-mentioned drawbacks of the prior art by providing a method and a system for quality control of objects in an apparatus for producing objects in a continuous cycle.

[0010] This object is fully achieved by the method and system of the present disclosure for quality control of objects in an apparatus for producing objects in a continuous cycle, as characterized in the appended claims.

[0011] According to one aspect thereof, the present disclosure provides a method for quality control of objects in an apparatus for producing objects in a continuous cycle. The method includes the step of individually supplying the objects to an inspection station. The method includes the step of capturing an image of each object disposed at the inspection station.

[0012] In one example, the image is taken by an optical device. The optical device may include a camera. The optical device may include an illuminator for irradiating the object within the inspection station. The optical device observes the object disposed at the inspection station. Thus, when the object disposed at the inspection station is irradiated by the illuminator, an image of the object is taken by the optical device (camera). Preferably, the illuminator irradiates the object with light in the visible light, or IR, or UV spectrum.

[0013] The method also includes applying a first processing step to each image. The first processing step is performed to assign the image and the corresponding object to one of two categories: a defective object category and a non-defective object category.

[0014] If the image belongs to the defective object category, the method is a step of applying a second processing step to the image data associated with the image, and includes further classifying the image and the corresponding object according to a plurality of defect categories. The step of classifying the image and the corresponding object is performed based on a plurality of identification features. In one example, the plurality of identification features are extracted from the image data. The plurality of identification features are extracted from the image data in real time or, alternatively, in post-processing. Thus, the image data is processed to extract a plurality of identification features.

[0015] This solution can separate defective objects from non-defective objects and further classify the defects in a particularly efficient manner.

[0016] It should be noted that the present disclosure also includes taking measures to adjust the production apparatus in response to the detected defects, thanks to the possibility of identifying various types of defects of the object. These measures can be either automatic or manual. In this way, the production apparatus may be provided with a feedback control system. For example, criteria based on the identification of defects (e.g., criteria including the avoidance of certain types of defects) can be used to update or adjust one or more control parameters (controlling the corresponding steps of continuous cycle production) and / or to update the settings of one or more components of the apparatus.

[0017] In one example, the optical device includes a camera. In one example, the image captured for each object represents the visible appearance of the object. The image is taken by the camera.

[0018] In one example, the method includes the step of storing in a database an image belonging to a defective object category. This solution enables having a database for reference, for example, during the classification step.

[0019] As a result of a process in which a plurality of identification features are extracted from each image data, an array is generated for each image data, and the array includes the values of the identification features for that image data. Such an array constitutes the fingerprint of the image data and thus of the corresponding object. The plurality of identification features define a working space, and each identification feature constitutes one dimension of the working space. Thus, the working space has a plurality of dimensions. Each dimension of the plurality of dimensions of the working space corresponds to one feature of the plurality of identification features extracted from the image data of each image. In particular, the values of the identification characteristics extracted for each image define the position of the image data of each image in the working space.

[0020] In one example, unsupervised clustering is used in a second processing step. During unsupervised clustering, each image data (image data related to an image captured for each object) can be represented as a (data) point in the working space (in fact, the array of that image data provides a plurality of coordinates in the working space). In one example, in unsupervised clustering, defective categories are generated by grouping data points having similar positions in the working space. This solution enables identifying various defective categories, including categories that were not considered before the start of quality control. Further, in the step of generating defective categories, the category with the largest number of defects can be identified.

[0021] "Unsupervised clustering" means a grouping system for subdividing data points in a working space in an unsupervised manner into groups.

[0022] It should be noted that classifying (or identifying) defects in accordance with the present disclosure is suitable for enabling the user to utilize the classification (or identification) results in a particularly simple and readable manner. For example, the output of unsupervised classification may be a report (or map) regarding various types of defects identified in an object (e.g., considering the population of objects). Thus, there is no need for a specialist to check the output, and even an operator outside the specialty can read the output and confirm the various types of defects and the number of defects in each defect category.

[0023] According to another aspect of the present disclosure, unsupervised classification (i.e., the step of clustering) can be started at any time. In this way, it is also possible to create a system of "continuous classification".

[0024] In practice, by unsupervised clustering, the classification of objects can be repeated whenever an object is identified as being defective, or at predetermined time intervals, or after a certain number of objects have been identified as being defective, or according to other predetermined criteria.

[0025] According to another aspect, if an object is classified as defective in a first processing step but is not recognized as belonging to one of the already identified defect categories in a second processing step, the system can create a new defect category (cluster) in the working space (thanks to unsupervised clustering). Thus, it is possible to add new defect categories and continuously update the existing categories (i.e., throughout the operation of the device).

[0026] In one example, images belonging to the non-defective object category are excluded from the storage step.

[0027] This solution makes it possible to use less database capacity. Furthermore, by not storing data related to non-defective objects, the quality control process is speeded up.

[0028] In one example, the first processing step provides location information. The location information is related to the location of the defect of each defective object.

[0029] In one example, the location information is supplied to a second processing step. This information can be used to classify the defects.

[0030] Furthermore, in one example, the plurality of identification features includes at least one feature representing the location information.

[0031] In one example, the first processing step is performed by a machine learning model. The machine learning model is trained to assign each image to a defective object category or a non-defective object category. The machine learning model is trained based on training data. The training data may include only images of non-defective objects. With this solution, the machine learning model can be trained using images of non-defective objects. Therefore, defects can be identified without the need for a complete database of defects.

[0032] Furthermore, the first processing step may include, for each image, extracting a diagnostic marker from the image data and applying a predetermined diagnostic rule (i.e., an algorithm).

[0033] In one example, the first processing step includes, in a first stage, a machine learning model. The machine learning model is trained to assign each image to a defective object category or a non-defective object category. The machine learning model is trained based on training data. In one example, the training data may include only images of non-defective objects. The first processing step may also include, in a second stage, for each image, extracting a diagnostic marker from the image data and applying a predetermined diagnostic rule (i.e., an algorithm). Furthermore, both the result of the first stage and the result of the second stage of the first processing step are considered to assign the image and the corresponding object to a defective object category or a non-defective object category.

[0034] In one example, in the second processing step, the output data of both the first and second stages of the first processing step are received and processed in combination with each other.

[0035] In one example, in the first processing step, both the first and second stages are applied to the image data obtained from each object.

[0036] In another example, the image data of each object may be divided into a first subset and a second subset according to a predetermined criterion in the first processing step. In this solution, for each object, the first stage is applied to the first subset and the second stage is applied to the second subset.

[0037] In one example, the first processing step may also include a plurality of tasks that provide a corresponding plurality of conditions related to the objects to be checked according to a predetermined sequence. In this solution, the first group of tasks is executed by a machine learning model, and the second group of tasks is executed by applying a predetermined diagnostic rule to extract diagnostic markers from the image data.

[0038] According to one aspect thereof, the present disclosure also provides a system for quality control of an object in an apparatus that produces an object in a continuous cycle. A system (hereinafter, the system) for quality control of an object in an apparatus that produces an object in a continuous cycle includes an optical device. The optical device is configured to capture an image of each object located at an inspection station. The system may include a conveyor. The conveyor is configured to individually supply the objects to the inspection station. The system also includes a processing unit. The processing unit is programmed to process each image in a first processing step. The processing unit is programmed to assign the image and the corresponding object to one of two categories, a defective object category and a non-defective object category.

[0039] In a second processing step, the processing unit is also configured to process the image data associated with each image belonging to the defective object category in response to the result of the first processing step, and classify the images and corresponding objects according to a plurality of defect categories. The second processing step is executed based on a plurality of identification features. In one example, the plurality of identification features are extracted from the image data.

[0040] In one example, the system includes a storage unit. The storage unit is configured to store images belonging to the defective object category in a database.

[0041] In one example, the processing unit is configured to perform unsupervised clustering in the second processing step. The unsupervised clustering is programmed to define a working space. The working space has a plurality of dimensions. Each dimension corresponds to one of the plurality of identification features extracted from the image data of each image. The values of the identification features extracted for each image define the position of the image data of each image in the working space, whereby each image data is represented as a data point in the working space. Thus, the unsupervised clustering is programmed to represent each image data as a data point in the working space. The unsupervised clustering is programmed to generate defect categories by grouping data points having similar positions in the working space.

[0042] In one example, the processing unit is configured to obtain position information related to the position of the defect in each defective object in the first processing step.

[0043] In the first processing step, the processing unit may include a machine learning model. The machine learning model is trained to assign each image to a defective object category or a non-defective object category. In one example, the machine learning model is trained based on training data. In one example, the training data may include only images of non-defective objects.

[0044] Furthermore, in the first processing step, the processing unit may include, for each image, extracting a diagnostic marker from the image data and applying a predetermined diagnostic rule (i.e., an algorithm).

[0045] In the first processing step, the processing unit may include a machine learning model in the first stage. The machine learning model is trained to assign each image to a defective object category or a non-defective object category. The machine learning model is trained based on training data. In one example, the training data may include only images of non-defective objects.

[0046] Furthermore, in the first processing step, the processing unit may include, in the second stage, extracting a diagnostic marker from the image data for each image and applying a predetermined diagnostic rule.

[0047] In one example, both the results of the first stage and the second stage of the first processing step are considered to assign the image and the corresponding object to a defective object category or a non-defective object category.

[0048] In one example, in the second processing step, the output data of both the first stage and the second stage of the first processing step are received and processed in combination with each other.

[0049] According to one aspect thereof, the present disclosure provides an apparatus for producing an object in a continuous cycle. The apparatus includes one or more machines for producing the object. The apparatus also includes a system for performing quality control of the object. The system for performing quality control of the object is made according to the present disclosure.

[0050] According to one aspect thereof, the present disclosure provides a computer program. The computer program includes instructions configured to perform quality control of an object in an apparatus for producing an object in a continuous cycle according to the present disclosure.

Brief Description of the Drawings

[0051] These features and other features will become more apparent from the following description of the preferred embodiments, which are illustrated as non-limiting examples in the accompanying drawings.

Figure 1

Figure 2

Figure 3

DETAILED DESCRIPTION OF THE INVENTION

[0052] Referring to the accompanying drawings, the numeral 1 denotes a system for quality control of an object O in an apparatus for producing an object O in a continuous cycle. The system 1 includes an optical device 101. The optical device 101 is configured to capture an image I of each object O located at the inspection station IP. In another example, the optical device 101 is configured to capture a plurality of images of each object. In one example, the optical device 101 includes a camera.

[0053] The optical device 101 includes an illuminator for irradiating an object within the inspection station. Preferably, the illuminator is configured to irradiate an object within the inspection station IP with light in the visible, IR, or UV spectrum. Further, the optical device includes a camera configured to observe an object within the inspection station.

[0054] In one example, the image I acquired for each object O placed at the inspection station represents the visible appearance of the object. In other words, in one example, each object is irradiated with light in the visible spectrum. In one example, the object may be exposed to infrared rays. According to one example, the object O within the inspection station is exposed to light in the visible, IR, or UV spectrum (more generally, light in a spectrum other than X-rays). In particular, the image obtained for each object shows the visible side of the object. The image obtained for each object is an image of the entire object (representing the entire object).

[0055] System 1 may also include a conveyor C. The conveyor C is configured to individually supply the object O to the inspection station IP. In other words, in a preferred embodiment, the objects are conveyed one by one to the inspection station at a time. In this solution, each object is transported to the inspection station in a predetermined orientation. Further, each object may be irradiated according to a predetermined orientation within the inspection station. In another example, the conveyor may be configured to supply the objects in a disorderly flow such that a plurality of objects are present at the inspection station at a time. Thus, the captured image of the objects within the inspection station may include a plurality of objects. The conveyor is configured to supply the object O in the supply direction F. The system also includes a processing unit 102. The processing unit 102 is programmed to process each image in a first processing step. The processing unit 102 is configured to assign the image I and the corresponding object O to one of two categories, a defective object category and a non-defective object category. The processing unit 102 is also configured to process the image data associated with each image I belonging to the defective object category in a second processing step. The processing unit 102 is configured to process the image data associated with each image I belonging to the defective object category in response to the result of the first processing step. In one example, the processing unit 102 is configured to process only the image data associated with the images belonging to the defective object category. The processing unit 102 is configured to perform a second processing step of classifying the images and the corresponding objects according to a plurality of defect categories. The processing unit 102 is configured to classify the images and the corresponding objects based on a plurality of identification features. In one example, the plurality of identification features are extracted from the image data.

[0056] In one example, the processing unit 102 includes a storage unit. The storage unit 1021 is configured to store the images belonging to the defective object category in a diagnostic database. Further, in one example, the images belonging to the non-defective object category are removed. Preferably, the storage unit includes a non-volatile memory.

[0057] In one example, in the first processing step, the processing unit is configured to obtain position information related to the position of the defect of each defective object O.

[0058] In one example, in the first processing step 102A, the processing unit 102 includes a machine learning model. The machine learning model is trained to assign each image to a defective object category or a non-defective object category. The machine learning model is trained based on training data. In one example, the training data includes only images of non-defective objects.

[0059] Furthermore, in the first processing step, the processing unit includes extracting a diagnostic marker from the image data for each image. In the first processing step, the processing unit includes applying a predetermined diagnostic rule to each image.

[0060] More specifically, each image of the object to be inspected provides a set of pixels, and the defect can take the form of an inconsistent area, such as a different value of luminosity or color in an area (non-defective area) that contrasts with it. The contrast value depends on the variation in luminosity compared to non-defective objects and generally varies at each point in the image of the object to be inspected. This definition applies to defects that take the form of areas of uniform color.

[0061] In the case of a texture area, the defect can take the form of an area of pixels that includes a variation (i.e., contrast) compared to the normal variation of the shade or color (including positional variation) of non-defective objects. Therefore, a change in texture can be regarded as a defect.

[0062] In one example, in the first processing step 102A, the processing unit 102 includes a machine learning model in a first stage. The machine learning model is trained to assign each image to a defective object category or a non-defective object category. The machine learning model is trained based on training data. The training data includes only images of non-defective objects.

[0063] Furthermore, in one example, in the second processing step 102B, the processing unit 102 includes, at the second stage, for each image, extracting a diagnostic marker from the image and applying a predetermined diagnostic rule.

[0064] More specifically, both the result of the first stage and the result of the second stage of the first processing step are considered in order to assign the image and the corresponding object to a defective object category or a non-defective object category.

[0065] In one example, the processing unit 102 is configured to perform unsupervised clustering in the second processing step 102B. The unsupervised clustering is programmed to represent each image data as a data point in a working space. In particular, the working space has a plurality of dimensions. Each dimension corresponds to one of a plurality of identifying features. The values of the identifying features extracted for each image (providing an array that constitutes a plurality of coordinates in the working space for that image and thus identifying a point in the working space) define the position of the image data of that image in the working space, whereby each image data can be represented as a (data) point in the working space. The unsupervised clustering is programmed to generate a defective category by grouping data points having similar positions in the working space. Thus, according to one example, for each image obtained for each object, a plurality of identifying features are extracted from the image data of each image, and the value of each feature extracted from the image data of each image determines the position of the image data of that image in the working space.

[0066] More specifically, "unsupervised clustering" refers to a grouping system for subdividing data points in a working space in an unsupervised manner. In other words, an unsupervised grouping system (i.e., unsupervised clustering) divides data points in a working space based on the positional similarity of the data points, and as a result, creates various groups (or defect categories) of data points in the working space. Unsupervised clustering can also label each group of data points in the working space.

[0067] According to one aspect, the present disclosure provides a method for quality control of an object O in an apparatus for producing an object in a continuous cycle. The method includes the step of individually supplying the object O to an inspection station IP. In other words, in a preferred embodiment, the objects are transported to the inspection station one by one at a time. In this solution, each object is transported to the inspection station in a predetermined orientation. The method may also include the step of irradiating each object according to a predetermined direction within the inspection station. In another example, the method may include the step of supplying the objects in a disorderly flow such that a plurality of objects are present at the inspection station at a time. Thus, an image of the objects captured at the inspection station may include a plurality of objects. The method includes the step of capturing an image I of each object O disposed at the inspection station IP. The method also includes the step of applying a first processing step 102A to each image. The first processing step 102A is performed to assign the image and the corresponding object to one of two categories, a defective object category and a non-defective object category. The first processing step may include extracting features from the input data (i.e., the image). The extracted data is processed in the first processing step to detect defects. Further, if the image belongs to the defective object category, the method includes the step of applying a second processing step 102B to the image data. The image data is that of each image. The image data may be data from the raw image of the object. The image data may be semi-processed data obtained from the first processing step 102A. The second processing step 102B is performed to classify the image I and the corresponding object O according to a plurality of defect categories. The classification of the defects is performed based on a plurality of identifying features. In one example, the identifying features are extracted from the image I. In one example, the second processing step 102B is applied only to the objects belonging to the defective object category.

[0068] The method also includes the step of storing the images belonging to the defective object category in a (diagnostic) database. In one example, the images belonging to the non-defective object category are excluded from the storing step. Preferably, the images belonging to the non-defective object category are removed.

[0069] In one example, the first processing step 102A is performed by a machine learning model. The machine learning model is trained to assign each image to a defective object category or a non-defective object category. The machine learning model is trained based on training data. In one example, the training data may include only images of non-defective objects.

[0070] Furthermore, the first processing step 102A may include extracting a diagnostic marker from the data of the image I for each image I. The first processing step 102A may include applying a predetermined diagnostic rule. In one example, for each image, a map of diagnostic markers extracted from the image is obtained. A predetermined rule (or algorithm) is applied to the map to identify if there are defects in the image and the corresponding object.

[0071] For example, the defect appears in the form of a variation in the brightness of a part of the image, resulting in contrast in that part. Therefore, detecting this kind of difference on the map means that a defect has been detected. In one example, the first processing step 102A includes a machine learning model in the first stage 1021A. Furthermore, the step of extracting the diagnostic marker is performed in the second stage 1022A of the first processing step 102A. In one example, the processing steps of the first stage and the second stage are performed simultaneously.

[0072] Furthermore, during the first processing step 102A, position information related to the position of the defect in each defective object is obtained. The position information is supplied to the second processing step 102B. In one example, the plurality of identification features includes at least one feature representing the position information. In one example, the position information is obtained in the second stage of the first processing step 102A.

[0073] In one example, unsupervised clustering is used in the second processing step 102B. During unsupervised clustering, each image data is represented as a data point in the working space, and defect categories are generated by grouping data points having similar positions in the working space. In one example, if the system is unable to recognize at least two different data points in the working space, the system attempts to display a second set of data points to distinguish groups of data points. Through the interface, the user can also add new defect categories or modify (highlight, separate, label) already recognized defect categories. In one example, both the results of the first stage and the second stage of the first processing step 102A are considered to assign an image and the corresponding object to a defective object category or a non-defective object category.

[0074] In one example, in the second processing step, the output data of both the first stage and the second stage of the first processing step 102A are sent as input to the second processing step. More specifically, in the second processing step 102B, the output data of both the first stage and the second stage of the first processing step 102A are received and processed in combination with each other.

[0075] In one example, in a first processing step, both a first stage and a second stage are applied to each image data obtained from each object. Thus, each image obtained from each object can be checked by a machine learning model or by extracting diagnostic markers from the image data and applying predetermined diagnostic rules, and can be assigned to a defective object category or a non-defective object category. Further, in one example, the image data of each object may be divided into a first subset and a second subset according to a predetermined criterion in the first processing step. In this solution, for each object, the first stage is applied to the first subset and the second stage is applied to the second subset. For example, a predefined portion of the object is analyzed using a machine learning model, diagnostic markers are extracted from the image data, and another predefined portion is analyzed by applying predetermined diagnostic rules (e.g., an artificial neural network). Thus, according to a predetermined criterion, the image data of each object can be subdivided into a first subset and a second subset, and a combination of the first stage and the second stage can be applied to each object.

[0076] The first processing step may also include a plurality of tasks. The plurality of tasks can provide corresponding multiple conditions to be satisfied according to a predetermined sequence. The multiple conditions to be satisfied may relate to the object to be checked. In this solution, a first group of tasks may be executed by a machine learning model, or a second group of tasks may be executed by extracting diagnostic markers from the image data and applying predetermined diagnostic rules.

[0077] According to one aspect thereof, the present disclosure provides an apparatus for producing an object in a continuous cycle. The apparatus includes one or more machines for producing the object. The apparatus also includes a system 1 for performing quality control of the object, and the system 1 is according to the present disclosure.

[0078] According to another aspect thereof, the present disclosure provides a computer program. The computer program includes instructions configured to execute the steps of the method according to the present disclosure.

Prior Art Documents

Patent Documents

[0079]

Patent Document 1

Patent Document 2

Patent Document 3

Patent Document 4

Patent Document 5

Patent Document 6

Claims

1. A method for quality control of an object in an apparatus for producing the object in a continuous cycle, comprising: - supplying the object (O) individually to an inspection station (IP); - capturing an image (I) for each object (O) disposed at the inspection station (IP); - applying a first processing step (102A) to each image, the step of assigning the image and the corresponding object to one of two categories: a defective object category and a non-defective object category; - when the image belongs to the defective object category, applying a second processing step (102B) to the image data associated with the image, and further classifying the image and the corresponding object according to a plurality of defect categories based on a plurality of identification features extracted from the image data.

2. The method according to claim 1, wherein the optical device (101) irradiates the object in the inspection station (IP) with light in the visible light, IR, or UV spectrum, includes a camera, the camera observes the object, and takes a picture of the image (I) of the object.

3. The method according to claim 1 or 2, including the step of storing the image belonging to the defective object category in a database.

4. In the second processing step (102A), unsupervised clustering is used. During the unsupervised clustering, a working space having a plurality of dimensions is defined. Each dimension corresponds to one of the plurality of identification features extracted from the image data of each image. The values of the identification features extracted for each image define the position of the image data of each image in the working space, whereby each image data is represented as a data point in the working space. The defect category is generated by grouping data points having similar positions in the working space.

5. The method according to claim 3 or 4, wherein the image belonging to the non-defective object category is excluded from the storing step.

6. The method according to any one of claims 1 to 5, wherein position information related to the position of the defect in each defective object is obtained during the first processing step (102A).

7. The method according to claim 6, wherein the position information is supplied to the second processing step (102B). **Claim 8** The method according to claim 7, wherein the plurality of identification features includes at least one feature representing the position information. **Claim 9** The method according to any one of claims 1 to 8, wherein the first processing step (102A) is executed through a machine learning model trained to assign each image to the defective object category or the non-defective object category, and the machine learning model is trained based on training data including only images of non-defective objects. **Claim 10** The method according to any one of claims 1 to 9, wherein the first processing step (102A) includes extracting a diagnostic marker from the image data for each image and applying a predetermined diagnostic rule (algorithm). **Claim 11** The first processing step (102A) is - a machine learning model trained to assign each image to the defective object category or the non-defective object category, and - in a second stage (1022A), for each image, extracting a diagnostic marker from the image data and applying a predetermined diagnostic rule, and The method according to any one of claims 1 to 10, wherein both the result of the first stage and the result of the second stage of the first processing step are considered to assign the image and the corresponding object to the defective object category or the non-defective object category. **Claim 12** During the first processing step, one of the following conditions occurs, namely i) both the first stage and the second stage are applied to the image data obtained from each object, ii) according to a predetermined criterion, the image data of each object is divided into a first subset and a second subset, and for each object, the first stage is applied to the first subset and the second stage is applied to the second subset, iii) the first processing step includes a plurality of predetermined tasks, providing corresponding multiple conditions to be satisfied according to a predetermined sequence, a first task group of the plurality of tasks is executed by the machine learning model, and a second task group of the plurality of tasks is executed by extracting a diagnostic marker from the image data and applying a predetermined diagnostic rule, The method according to claim 11. **Claim 13** A system (1) for quality control of an object in an apparatus for producing an object in a continuous cycle, - an optical device (101) configured to capture an image (I) of each object (O) located at an inspection station (IP); - a conveyor (C) for individually supplying the objects (O) to the inspection station (IP); - a processing unit (102), which, in a first processing step (102A), processes each image (I) and assigns the image (I) and the corresponding object (O) to one of two categories, a defective object category and a non-defective object category; and, in a second processing step (102B), in response to the result of the first processing step (102A), processes the image data associated with each image belonging to the defective object category and classifies the image and the corresponding object according to a plurality of defect categories based on a plurality of identification features extracted from the image data. A system (1) comprising a processing unit (102) programmed as such.

14. The system (1) according to claim 13, further comprising a storage unit configured to store the images belonging to the defective object category in a database.

15. The processing unit (102) is configured to perform unsupervised clustering in the second processing step, the unsupervised clustering being programmed to define a working space of a plurality of dimensions, each dimension corresponding to one of the plurality of identification features extracted from the image data of each image, and the values of the identification features extracted for each image defining the position of the image data of each image in the working space, whereby each image data is represented as a data point in the working space, and the unsupervised clustering is programmed to generate the defect categories by grouping data points having similar positions in the working space. The system (1) according to claim 14.

16. The system (1) according to any one of claims 1 to 15, wherein the processing unit (102) is configured to obtain position information related to the position of a defect in each defective object in the first processing step (102A).

17. The processing unit (102) includes, in the first processing step (102A), a machine learning model trained to assign each image to the defective object category or the non-defective object category, the machine learning model being trained based on training data including only images of non-defective objects, the system (1) according to any one of claims 1 to 16.

18. The processing unit (102) includes, in the first processing step (102A), extracting a diagnostic marker from the image data for each image and applying a predetermined diagnostic rule, the system (1) according to any one of claims 1 to 17.

19. The processing unit (102) in the first processing step (102A), - a machine learning model trained to assign each image to the defective object category or the non-defective object category, and - in a second stage (1022A), for each image, extracting a diagnostic marker from the image data and applying a predetermined diagnostic rule, wherein both the result of the first stage of the first processing step and the result of the second stage are considered to assign the image and the corresponding object to the defective object category or the non-defective object category, the system (1) according to any one of claims 1 to 18.

20. The optical device (101) includes an illuminator that irradiates the object within the inspection station (IP) with light in the visible light, or IR, or UV spectrum, and a camera configured to observe the object in the inspection station (IP) and capture the image (I) of the object, the system (1) according to any one of claims 1 to 19.

21. An apparatus for manufacturing production in continuous cycles, - one or more machines for producing the object, and - a system (1) for performing quality control of the object, the system (1) being the system according to any one of claims 13 to 20, an apparatus comprising the system (1).

22. A computer program including instructions configured to execute the steps of the method according to any one of claims 1 to 12 when executed on a processor.

Images