DE102023126786B3 - Method and system for quality control of ophthalmic lenses - Google Patents

Abstract

The invention relates to a method for quality control of ophthalmic lenses (1) in a manufacturing system (100) comprising a plurality of successive stations (110, 120, 130, 140) for processing the lenses (1), wherein the lenses (1) are examined for quality control by means of at least a first control station (10) by generating a basic image (11) of the ophthalmic lens (1) in which at least one possible abnormality (A1) on the ophthalmic lens (1) is identified, wherein the method comprises at least the following steps: providing a first training database (12) in which labels (E) are stored which have information comprising various first abnormalities (A1); providing a first classification unit (13) in connection with the first training database (12); providing a first decision unit (14) in connection with the first control station (10) and in connection with the classification unit (13); Classifying the first abnormality (A1) of the basic image (11) on the ophthalmic lens (1) and assigning a label (E) from the first training database (12) to the ophthalmic lens (1) by means of a classification function (KF); generating a first result value (E1) from the classification of the first abnormality (A1) comprising the information of a positive evaluation, a negative evaluation or a rework evaluation; ...

Description

The invention relates to a method for quality control of ophthalmic lenses in a production system comprising several successive stations for processing the lenses, wherein the lenses are examined for quality control by means of at least one first control station by generating a basic image of the ophthalmic lens in which at least one possible abnormality on the ophthalmic lens is identified. The invention is also directed to a system for quality control of ophthalmic lenses by means of such a method.

STATE OF THE ART

In the WO 2023/041659 A1 A method for quality control of ophthalmic lenses is described. The quality control described here concerns the identification and classification of possible abnormalities on the ophthalmic lens, whereby lenses without abnormalities are also checked, for example as part of a 100% quality control. In this respect, the abnormalities are "possibly" present as stated, but they do not necessarily have to be present on every ophthalmic lens.

The EP 1 061 571 A2 discloses a method and apparatus for automatically classifying a defect on the surface of a semiconductor wafer into one of a predetermined number of core classes using a core classifier that uses boundary and topographical information. Defects that cannot be classified by either the core classifier or the specific adaptive classifiers are classified by a full classifier.

The WO 2023 / 007 110 A1 discloses a system and method for capturing and inspecting lens images of ophthalmic lenses using one or more cameras to capture the images of the lenses in a dry or wet state. The images are further processed and then input to an artificial intelligence network, such as a convolutional neural network (CNN), to analyze and characterize the nature of the lens defects. The artificial intelligence network identifies defect areas on the images and outputs defect categories or classifications for each of the images based in part on the defect areas.

The WO 2022 / 081 370 A1 relates to inline measurement systems, devices and methods for measuring optical devices and discloses examples of optical devices for augmented, virtual and/or mixed reality applications. A measurement system for optical devices is to be configured to measure a plurality of first metrics and one or more second metrics for optical devices, wherein the one or more second metrics comprise a so-called display leakage metric.

Possible abnormalities in the basic images recorded by the ophthalmic lenses can concern defects, but also intentional structures such as markings or the like, for example structures present in the mold, as well as contamination, which can be more or less pronounced depending on the abnormality.

The measurement method for quality control includes a light source for generating a sinusoidal stripe pattern, for example, which is recorded with a camera, whereby the light generated with the light source is passed through the ophthalmic lens and then recorded with a camera. The sinusoidal stripe patterns are then displayed on a screen, for example, and/or evaluated using an appropriate device. The sinusoidal stripe patterns can have phase shifts, have different wavelengths and run horizontally and vertically, for example. Using other known methods, certain abnormalities can be made more visible. The basic images generated are then stored in a database. The different types of abnormality are classified so that a decision can be made as to which class the abnormality should be assigned.

A disadvantage of quality control in a production system with several stations for processing and treating a large number of lenses arises when quality control is carried out by means of the control station before or after a first or a number of stations in the production system, whereby the processing or treatment of the ophthalmic lenses after further stations in the production system can either weaken or intensify the abnormalities. It can happen that certain abnormalities in the result from the first control station lead to the ophthalmic lens not being fed to the further stations in the production system at all, even though this would actually be possible because, for example, the abnormality could be eliminated by subsequent coating of the surface of the ophthalmic lenses or it could then at least be classified as a good part.

On the other hand, if certain abnormalities which are to be classified as minor are assigned to a category according to which the lens is sent to the further process, this can lead to the lens having to be classified as a defective part, contrary to the assessment from the first assessment unit, particularly after a coating or other type of surface treatment, even though simple rework, e.g. by the stations before the first inspection station, could ultimately have produced the lens as a good part. A separate manual station may also be set up, or the manually operated station may not be set up in a uniform manner with the station directly before the station which usually cleans the ophthalmic lenses. However, not every lens rated as defective from the second inspection station can be sent back for rework; this is particularly because once a coating has been applied, the lens cannot generally be returned to any previous station in the production system.

It is therefore desirable to improve the method for carrying out quality control so that the assessment of a good part of the ophthalmic lens in the first inspection station corresponds as closely as possible to the assessment of a good part after the entire treatment of the lens using the manufacturing system. In this respect, the deviation between the assessment from the first inspection station and the final assessment after further treatment of the lens through the entire manufacturing system should be as small as possible.

DISCLOSURE OF THE INVENTION

The object of the invention is to improve a method for quality control of ophthalmic lenses with at least one control station, wherein the method is to be operated in such a way that the result of the evaluation of the ophthalmic lens from the quality control of the first control station enables the most accurate prediction possible about the result of the quality of the ophthalmic lens after passing through the entire production system. The evaluation, classification and/or categorization of the abnormalities of the ophthalmic lens from the control of the first control station should therefore be as accurate as possible.

This object is achieved on the basis of a method according to claim 1 and on the basis of a system according to claim 12 with the respective characterizing features. Advantageous developments of the invention are specified in the dependent claims.

To solve the problem preceding the invention, the following further steps are proposed for the method: Providing a first training database in which labels are stored that have information comprising various first abnormalities; Providing a first classification unit in connection with the first training database, wherein the classification unit is trained by training with the training database; Providing a first decision unit in connection with the first control station and in connection with the classification unit; Classifying the first abnormality of the basic image on the ophthalmic lens and assigning a label to the ophthalmic lens by means of a classification function; Generating a first result value from the classification of the first abnormality comprising the information of a positive evaluation, a negative evaluation or a rework evaluation; Providing at least one second control station after the first, the second or a further station of the production system; Providing a second training database in which labels are stored that have information comprising various second abnormalities; Providing a second classification unit in connection with the second training database; Providing a second decision unit in connection with the second control station and in connection with the second classification unit; generating a second result value from the classification of the second abnormality comprising the information of a positive assessment, a negative assessment or a rework assessment; comparing the first result value with the second result value; generating feedback information from the comparison and introducing the feedback information into the first training database and into the classification function.

The basic image is a processed basic image, and from the sinus images, normally 24 to 100 individual images, several basic images are generated, e.g. 4 basic images. These 4 basic images are then further processed, various filters can be used, and in this way around 30 images can be generated in which the various abnormalities emerge, or the abnormalities can be determined from the combination of the 30 images.

Regarding the labels, it should be noted that a label can be assigned to each lens, but it is preferably provided that a label can be assigned to each abnormality, certain characteristics of the abnormality or even some or every pixel of an abnormality from the evaluation. In this way, each pixel can be assigned a label for in- Class Not-In-Class can be assigned for each class of abnormalities, also comprising an assessment of the strength of an abnormality, which then refers to contiguous regions of a class.

The result can also depend on the interaction of all abnormalities. Although a single abnormality can, for example, lead to a negative rating, it is also possible that this only leads to a negative rating in combination with another abnormality that would not have led to a negative rating on its own.

The core idea of the invention is to carry out the method for quality control of ophthalmic lenses with feedback between the second control station in conjunction with the second decision unit in order to make the result from the second decision unit available to the first control station with the first decision unit. This feedback system expands the method in a learning manner so that the quality of the information in the training database, in particular the labels and the classification thereof, in conjunction with the classification function, is constantly improved in the further application of the method.

The indication that the decision unit, the classification unit, the training database and/or the control station are connected to each other concerns an exchange of data between those entities.

With the at least two control stations according to the invention, a test infrastructure can be created with which data from the first control station can also be available at a later point in time, and the classification of an anomaly can also be called up at a later point in time, with measurement data being saved in order to check them later, but not all measurement data are later transferred to the training database. In the event of deviations in the assessment of the ophthalmic lens in the second control station due to defects that were already present when the first control station was checked, this deviation can be optimized by adapting the evaluation algorithm, with the evaluation algorithm being used, for example, through the classification function to classify the anomaly in the first control station.

In this way, the classification function can be trained in such a way that it is improved so that there are as few deviations as possible between the assessment of the abnormality from the first control station and that from the second control station. Storing the information in the training database in particular improves this further. The training database can receive additional information or existing information can be overwritten and thus improved in quality so that the classification unit classifies the abnormalities in the ophthalmic lenses better in order to obtain the same result from the second control station as the result from the first control station, so that the two result values match. The newly acquired information can be stored using labels that are overwritten or newly stored in the training database.

Thus, an essential feature of the invention is that the feedback information is used to train the classification function and/or create and/or improve the labels from the training database in such a way that the decision of the first decision unit as to whether the ophthalmic lens is a good part, a bad part, or a reworkable part is improved. For example, the first decision unit can classify an ophthalmic lens as a good part, but the assessed abnormality is such that after passing through the other stations of the production system, the classification of the second decision unit after the measurement by the second control station gives the ophthalmic lens a negative rating. The negative rating from the second decision unit is compared with the rating of the lens as a good part from the first decision unit, so that the labeling can consequently be carried out that precisely this abnormality is more likely to produce a bad part or at least a reworkable part in the future, i.e. for later, subsequent measurements. It should be noted that the reclassification by the first decision-making unit does not have to take place immediately; for example, the label can be rewritten or newly created only after an accumulation of a corresponding deviation with regard to a certain anomaly, in order to decide differently when the corresponding anomaly is repeated by the first decision-making unit.

The second result value from the second decision unit categorizes the lens as a good part, a bad part or a reworkable part. A good part either results from a good part of the first decision unit, resulting in a true positive evaluation, but the good part can also result from a bad part of the first decision unit, resulting in a false negative evaluation. A bad part can either result from a good part of the first decision unit that is incorrectly classified was, resulting in a false-positive assessment, or the bad part can result from a bad part from the first decision unit that is nevertheless fed into further treatment by the production system, resulting in a true-negative assessment. The bad part from the first decision unit can also be fed manually to the other stations in the production system, since it cannot be ruled out that the initially negative assessment of the abnormality from the first decision unit will ultimately result in a good part from the second decision unit due to the other stations, in particular due to a coating or other surface treatment, since the additional processing and treatment steps of the ophthalmic lenses in the other stations after the first control station reduce the abnormality or can even eliminate it completely. The assessment of the abnormalities refers to the result of the first station. Finally, it can also happen that the lens was classified as a good part in the first control station and a new defect was added for the first time in the further process, which leads to the lens being classified as a bad part in the second control station.

The classification function of the first decision unit is adjusted according to a particular advantage so that there is a minimal deviation of the first result value from the second result value.

In the first decision unit, a comparison is made between the abnormality detected by the first control station and an abnormality that is stored in the first training database, so that the abnormality is assigned to a specific classification, which is done with the classification unit. If the detected abnormality corresponds to a specific abnormality stored in the training database, the ophthalmic lens is classified accordingly and provided with a label, which label includes the classification data, wherein the label can also include metadata, in particular in order to be able to identify the lens with the corresponding abnormality during the subsequent run through the stations of the production system, in particular by the subsequently set up second classification unit or second decision unit.

The second anomaly, which is still present after passing through the stations of the production system, is labeled with the second classification unit so that the second decision unit can output a second result value accordingly. If the comparison of the second anomaly with a label from the second training database, for example, shows a very high level of agreement, a check is carried out to determine how the second anomaly should be categorized so that it is finally determined whether it is a good part, a bad part or a part that can be reworked, since lenses that have been passed through the coating stations can also be reworked if they have certain anomalies.

The classification function of the first decision unit is changed in such a way that the negative case of the evaluation decreases proportionately and subsequently more ophthalmic lenses are fed to the second station or other stations of the production system until the proportion of bad parts from the second result value increases and this increase, via the feedback information, counteracts the change in the classification function of the first decision unit. In other words, the change in the classification function can mean an overshoot of the second result value, so that the proportion of bad parts increases again, so that the classification function can be optimized so that the number of bad parts alternates around an optimum. Too much can therefore be sorted out in the first decision unit by making false-negative evaluations, which, however, does not allow an approximation to a real scrap rate without feedback. Likewise, too few ophthalmic lenses can be sorted out, which leads to a false-positive evaluation, which also causes the evaluated scrap rate to deviate further from the real scrap rate. However, with the method according to the invention, the number of lenses that are too many to be rejected, i.e. the false negative evaluation, can be reduced, as can the number of lenses that are too few to be rejected, which corresponds to the false positive evaluation. In this way, the evaluated rejection rate can be brought closer to the actual rejection rate, so that fewer lenses result from the second decision unit as defective parts or reworkable parts. The quality of the inspection is thus improved and the second result value results in a higher proportion of positively evaluated ophthalmic lenses.

As already mentioned, it is possible that ophthalmic lenses that have a negative evaluation from the first decision unit are subsequently evaluated as positive and are manually passed on to the next station. These are, for example, lenses that are categorised by the decision unit as rejects or not reworkable, but are nevertheless fed by an operator to the other stations of the production system. It can happen that, in particular, due to the influence on the Lens in the other stations, for example through a coating or other surface treatment, the abnormality is given a new category, according to which a positive assessment follows by the second decision unit. The ophthalmic lenses manually fed to the other station by a person are assessed using the second decision unit, with the resulting second result value of these lenses also being fed back to the classification function of the first decision unit via the feedback information. In particular, this allows the assessed rejection rate to be brought even closer to the actual rejection rate.

The first decision unit can work with the assistance of a person or completely autonomously; preferably, it can be operated purely by computer numerics. In contrast, the second decision unit can also be operated fully automatically, by computer numerics; preferably, the second decision unit is operated manually under further supervision by a person in addition to the computer numerics operation. It is also conceivable that the first decision unit and/or the second decision unit are operated by means of an AI module, i.e. comprising artificial intelligence, in particular for training the classification function based on the feedback information. The artificial intelligence can in particular help to optimize the quality of the labels from a large number of measurement data and/or a large number of different anomalies, so that the first classification unit creates a basis for the decision unit to better match the anomalies with the stored labels in the training database and to optimize the first decision so that the quality of the first result value of the first decision unit continues to increase.

The ophthalmic lenses are provided with the label either directly or at least indirectly, for example in conjunction with an identification number for the passage of the lens through the production system, whereby the labels contain at least data for the recognition of the specific ophthalmic lens and data relating to at least one of the first and second abnormalities or result values. In particular, so-called metadata can be part of the label in order to identify the lens at a later point in time and to retrieve the result values from the decision units.

The object of the invention is further achieved by a system for quality control of ophthalmic lenses, which is designed for use in a manufacturing system comprising several consecutive stations for processing the lenses. It is provided that a control station and a first decision unit are set up, and independently of this, at least a second control station and a second decision unit are set up, wherein at least the second decision unit is connected to the first decision unit via a communication channel, in particular a cloud system, via which feedback information can be transmitted from the second decision unit to the first decision unit in order to influence a classification function of the first decision unit. The further features and advantages described in connection with the method also apply to the system according to the invention for quality control of ophthalmic lenses.

In particular, at least the first decision unit and/or the second decision unit can comprise an AI module in order to use artificial intelligence; it is also conceivable that the first decision unit can be operated computer-numerically and that the second decision unit can be operated computer-numerically and/or manually by a person.

PREFERRED EMBODIMENT OF THE INVENTION

• 1 eine schematische Ansicht der Architektur des Systems zur Durchführung des erfindungsgemäßen Verfahrens zur Qualitätskontrolle ophthalmischer Linsen;
• 2 ein Diagramm, in dem die Entscheidungseinheiten zentral dargestellt sind, die von den ophthalmischen Linsen mit schematisch dargestellten Auffälligkeiten durchlaufen werden;
• 3 ein Diagramm, in dem eine Ausschussrate über der Zeit dargestellt ist, ohne dass es zu einer Rückkopplung zwischen der zweiten Entscheidungseinheit und der ersten Entscheidungseinheit kommt; und
• 4 ein Diagramm, in dem die Ausschussrate über der Zeit mit einer erfindungsgemäßen Rückkopplung zwischen der zweiten Entscheidungseinheit und der ersten Entscheidungseinheit.

Further measures improving the invention are described in more detail below together with the description of a preferred embodiment of the invention with reference to the figures. It shows:

• 1 a schematic view of the architecture of the system for carrying out the method according to the invention for quality control of ophthalmic lenses;
• 2 a diagram in which the decision units are centrally shown through which the ophthalmic lenses pass with schematically shown abnormalities;
• 3 a diagram showing a rejection rate over time without any feedback between the second decision unit and the first decision unit; and
• 4 a diagram in which the reject rate over time with an inventive feedback between the second decision unit and the first decision unit.

1 shows schematically several stations 110, 120, 130 and 140, which are shown only as examples and which Components of the manufacturing system 100 for processing and treating ophthalmic lenses 1 are shown. A series of ophthalmic lenses 1 is shown, which are fed to the first station 110, and in the stations 110 and 120 the lenses 1 are milled, turned and polished, for example. Laser engraving or other marking processes can also be carried out, which are also later subjected to quality control.

The second station 120 is followed by the first control station 10, which is designed, for example, according to the disclosure of WO 2023/041659 A1 .

The further stations 130 and 140 can, for example, carry out coating processes, hardening processes as well as other finishing of the surfaces of the ophthalmic lenses 1, so that a final inspection by means of the second inspection station 15 is only carried out after these treatment steps.

A basic image 11 emerges from the first control station 10 and is fed to the first classification unit 13 and the first decision unit 14. The first classification unit 13 in conjunction with the first decision unit 14 determines whether the ophthalmic lens 1 currently being tested is a positive case P or a negative case N. If the case P is positive, the lens 1 is passed on to the third station 130 shown as an example. If the case is negative, the ophthalmic lens 1 is either completely sorted out, shown with a crossed-out lens, and the further negative case can mean that the lens is fed back to a previous station 110 and/or 120 or is manually reworked if the ophthalmic lens 1 can be reworked. In the first decision unit 14, a first result value E1 is generated on the basis of which the decision is made as to whether the lens represents a positive case P or a negative case N. For this purpose, the first decision unit 14 can have an AI module, shown as a separate unit by way of example, but which can also be part of the decision unit 14.

The second control station 15 is connected to a second classification unit 17 and a second decision unit 18, which checks the further processed ophthalmic lens 1 and produces a second result value E2, on the basis of which it is decided whether it is a good part or a bad part. The second training database and the second training unit are not shown here.

The second result value E2 is transmitted via feedback information R, in particular using a cloud system 19, first to the first training database 12 and to the first training unit 21, so that the parameters PA can be trained therein, with which the classification function KF is supplied.

The classification function KF is coupled to the first decision unit 14, so that the first decision unit 14 can ultimately determine, by evaluating the parameters PA from the training unit 21, whether the first result value E1 and the second result value E2 match. If these match, shown with a comparison V, the decision from the first decision unit 14 was correct; if they do not match, the decision from the first decision unit 14 was incorrect. The feedback information R thus continues from the second result value E2 via the cloud system 19, the first training database 12 and the first training unit 21, which can be AI-based from an AI module, to the classification function KF.

The feedback is made up of the comparison of the result values E1 and E2, which identifies that there is potential for improvement. This allows new labels E to be generated from the classified measurement data from the control stations 10 and the training database 12, which in turn are entered into the training database 12. New parameters PA are then generated via the training module 21 and passed on to the classification unit 13 or entered into the training database.

In order to generate the case in which an ophthalmic lens 1 that was classified as a negative case N by the first decision unit 14 is nevertheless fed to the further process, i.e. to the third station 130, a manual feed MZ can be carried out so that an operator still feeds the negatively assessed ophthalmic lens 1 to the third station 130. If the classification of this ophthalmic lens 1 in the second control station 15 in conjunction with the second decision unit 18 is positive, a false-negative assessment is produced. In the result from the second decision unit 18, there are therefore good parts and bad parts, i.e. lenses 1 that can be used and lenses 1 that are either rejects or could be reworked. The good parts can therefore result from a true-positive assessment TP, a false-negative assessment FN, and the bad parts can result from a false-positive assessment FP or a true-negative assessment TN.

The illustration also shows a cloud system 19, via which communication between the second decision unit 18 and the first decision unit 14 can take place even over greater distances. Of course, direct connections via a cable system, a radio system or the like are also possible, especially if the first and second control stations 10, 15 or the first and second decision units 14, 18 are not far from each other. In this respect, a local computer can also be used. However, if, for example, the second control with the second control station 15 takes place remotely from the first control with the first control station 10, for example if the production system 100 is designed in several parts and positioned at different locations, the cloud system 19 can be correspondingly advantageous.

2 shows in a schematic view first the first decision unit 14 and the second decision unit 18, wherein the first decision unit 14 is designed purely computer-numerically, while the second decision unit 18 can be computer-numerically or equipped with an operator.

The lenses 1 are handed over to the first inspection station 10, each with a first abnormality A1, for example. The abnormalities A1 can represent defects, intentional structures, markings, structures from the mold, contamination and the like, whereby the defects in particular can be divided into very many different classifications, for example they can be evaluated based on the quotient of the circumference and the area of the defects that are detected using the measuring method from the first inspection station 10, which is located upstream of the first classification unit 13 and the first decision unit 14. The same applies to the second inspection station 15, which checks the finished lenses 1 in order to finally feed the result to the second classification unit 18 and the second decision unit 18.

This large number of different designs of the first and second abnormalities are stored in respective training databases 12 and 16 as labels E. The first training database 12 stores a correspondingly large number of variously executed first and second abnormalities in the form of labels E, which are stored as a type of comparison value in the training database 12 or 16. If an ophthalmic lens 1 is now to be classified with a first abnormality A1, the abnormality A1 is considered via the first classification unit 13. The first training database 12 is followed by a first training unit 21 in which parameters PA are generated, based on which the first classification unit 13 classifies the lenses 1 to be tested with the specific abnormalities A1. A training procedure must be distinguished from the actual process. During training, it is assessed what similarities the abnormalities A1 have with the parameters stored in the labels E. During the process itself, it is then checked whether the abnormalities correspond to these parameters and they are labeled accordingly.

Finally, these lenses 1 are fed for further processing with the respective case evaluations as positive case P or negative case N, represented by the production system 100 following the first decision unit 14. The further processing in the production system 100 results in further processed ophthalmic lenses 1, which are passed on to the second control station 15, which is located upstream of the second classification unit 17 and the second decision unit 18. The second decision unit 18 is also connected to the second training database 16 storing the labels E via the second classification unit 17. Parameters PA from the second training unit 22 are passed on to the second classification unit 17 in the same way, so that the comparison process is repeated in order to finally classify the lenses 1 accordingly from the second decision unit 18, wherein it is shown by way of example that the middle lens now also represents a positive case P, but was classified as a negative case N from the first decision unit 14. It is also possible to feed the lenses 1 to the production system 100 again from the second control station 15 and the second decision unit 18 in order to enable rework.

The core of the invention is now the comparison between the result of the first decision unit 14 and the second decision unit 18 with the respective result values E1 and E2, so that feedback information R is now generated from this comparison, which is transferred to the first training database 12 or the first classification unit 13. This makes it possible to improve the quality of the database using the first anomalies A1 and the corresponding classifications, and it is also possible to store further anomalies A1 in the first training database 12.

Using an artificial intelligence AI (not shown in detail here), the first classification unit 13 can be executed as a learning system over the further operating period of the control system in order to continue to classify the abnormalities A1 of the ophthalmic lenses 1 by means of the first classification unit 13, in particular to improve the quality of the label E which is attached to the lenses 1.

The 3 and 4 show a so-called normalized reject rate nA over time t. The curve represents the evaluated reject rate bA, which alternates with a real reject rate rA. The curve of the evaluated reject rate bA shows that the evaluated reject rate bA according to 3 without feedback information between the second decision unit and the first decision unit does not approximate the real reject rate rA, while according to 4 with feedback information between the second and the first decision unit, the assessed rejection rate bA approaches the real rejection rate rA. The false-negative evaluations are plotted above the real rejection rate rA, and the false-positive evaluations are plotted below. Through the feedback information, which is calculated according to 4 is applied, the first decision unit in conjunction with the classification unit learns the improved evaluation, so that the result value of the second decision unit more often agrees with the result value of the first decision unit, shown by an estimated scrap rate bA approaching the real scrap rate rA.

The invention is not limited in its embodiments to the preferred embodiment given above. Rather, a number of variants are conceivable which make use of the solution presented even in fundamentally different embodiments. All features and/or advantages arising from the claims, the description or the drawings, including structural details or spatial arrangements, can be essential to the invention both individually and in a wide variety of combinations.

List of reference symbols:

1

ophthalmic lens

10

first checkpoint

11

basic image

12

first training database

13

first classification unit

14

first decision-making unit

15

second control station

16

second training database

17

second classification unit

18

second decision-making unit

19

cloud system

100

manufacturing system

110

first stop

120

second station

130

third station

140

fourth station

E

label

E1

first result value

E2

second result value

PA

parameter

P

positive case

N

negative case

FN

false-negative evaluation

TP

true-positive evaluation

FP

false-positive evaluation

TN

true-negative evaluation

R

feedback information

V

Comparison

A1

first abnormality

A2

second abnormality

KF

classification function

MZ

manual feeding

n/a

normalized scrap rate

rA

real scrap rate

bA

assessed scrap rate

Claims (13)

Method for quality control of ophthalmic lenses (1) in a production system (100) comprising a plurality of successive stations (110, 120, 130, 140) for processing the lenses (1), wherein the lenses (1) are examined for quality control by means of at least one first control station (10) by generating a basic image (11) of the ophthalmic lens (1) in which at least one possible abnormality (A1) on the ophthalmic lens (1) is identified, wherein the method comprises at least the following steps: - providing a first training database (12) in which labels (E) are stored which have information comprising various first abnormalities (A1); - providing a first classification unit (13) in connection with the first training database (12); - providing a first decision unit (14) in connection with the first control station (10) and in connection with the classification unit (13); - Classifying the first abnormality (A1) of the basic image (11) on the ophthalmic lens (1) and assigning a label (E) based on the parameters of the training from the first training unit to the ophthalmic lens (1) by means of a classification function (KF); - Generating a first result value (E1) from the classification of the first abnormality (A1) comprising the information of a positive evaluation, a negative evaluation or a rework evaluation; - Providing at least one second control station (15) after the first (110), the second (120) or a further station (140) of the production system (100); - Providing a second training database (16) in which labels (E) are stored which have information comprising various second abnormalities (A2); - Providing a second classification unit (17) in connection with the second training database (16); - Providing a second decision unit (18) in connection with the second control station (15) and in connection with the second classification unit (17); - Generating a second result value (E2) from the classification of the second abnormality (A2) comprising the information of a positive assessment, a negative assessment or a rework assessment; - Comparing the first result value (E1) with the second result value (E2); - Generating feedback information (R) from the comparison and - introducing the feedback information (R) into the first training database (12) and into the classification function (KF). procedure according to claim 1 , characterized in that the training of the classification function (KF) and/or the creation and/or improvement of the labels (E) from the training database (12) is carried out by means of the feedback information (R), such that the decision of the first decision unit (14) is improved as to whether the ophthalmic lens (1) is present as a good part (G), as a bad part (S) or as a reworkable part (N). procedure according to claim 1 or 2 , characterized in that the second result value (E2) the ophthalmic lens (1) is present as a good part (G), as a bad part (S) or as a reworkable part (N), wherein - a good part (G) either results from a good part (G) of the first decision unit (14), so that a true-positive assessment (TP) is produced, or that the good part (G) results from a bad part (S) of the first decision unit (14), so that a false-negative assessment (FN) is produced and/or wherein - the bad part (S) either results from a good part (G) of the first decision unit (14), so that a false-positive assessment (FP) is produced, or the bad part (S) results from a bad part (S) of the first decision unit (14), so that a true-negative assessment (TN) is produced. Method according to one of the preceding claims, characterized in that the classification function (KF) of the first decision unit (14) is adapted via the feedback information (R) such that a minimal deviation of the first result value (E1) from the second result value (E2) results. Method according to one of the preceding claims, characterized in that the classification function (KF) of the first decision unit (14) is changed such that the negative case (N) of the evaluation decreases proportionately and subsequently more lenses (1) are fed to the second station (120) or further station of the production system (100) until the proportion of bad parts (S) from the second result value (E2) increases and this increase, via the feedback information (R), adjusts the classification function (KF) of the first decision unit (14) again to counteract the change. Method according to one of the preceding claims, characterized in that ophthalmic lenses (1) which have the negative case (N) of the evaluation from the first decision unit (14) are subsequently evaluated as positive and are manually transferred to the further station (130). procedure according to claim 6 , characterized in that the ophthalmic lenses (1) manually fed to the further station (130) by a person are evaluated by means of the second decision unit (18), wherein the resulting second result value (E2) of these lenses (1) is fed back to the classification function (KF) of the first decision unit (14) via the feedback information (R). Method according to one of the preceding claims, characterized in that a local data network or a cloud system (19) is set up for data exchange, wherein the first control station (10), the second control station (15), the first decision unit (14) and/or the second decision unit (18) are communicated with each other by means of the cloud system (19), in particular in a decentralized manner. communicate with each other or exchange data. Method according to one of the preceding claims, characterized in that the first decision unit (14) is operated computer-numerically and that the second decision unit (18) is operated computer-numerically and/or manually by means of a person. Method according to one of the preceding claims, characterized in that the first decision unit (14) and/or the second decision unit (18) is operated by means of an AI module comprising artificial intelligence, in particular for training the classification function (KF) based on the feedback information (R). Method according to one of the preceding claims, characterized in that the ophthalmic lenses (1) are provided directly or at least indirectly with a label (E) which has at least data for the recognition of the specific ophthalmic lens (1) and data relating to at least one of the first and second abnormalities (A1, A2). System for quality control of ophthalmic lenses (1), which is designed for use in a manufacturing system (100) comprising a plurality of successive stations (110, 120, 130, 140) for processing the lenses (1), characterized in that a first control station (10) and a first decision unit (14) are set up and, independently thereof, at least a second control station (15) and a second decision unit (18) are set up, wherein at least the second decision unit (18) is connected to the first decision unit (14) via a communication channel, in particular a cloud system (19), via which feedback information (R) can be transmitted from the second decision unit (18) to the first decision unit (14) in order to influence a classification function (KF) of the first decision unit (14). system according to claim 12 , characterized in that at least the first decision unit (14) and/or the second decision unit (18) has an AI module comprising artificial intelligence and/or that the first decision unit (14) can be operated computer-numerically and that the second decision unit (18) can be operated computer-numerically and/or manually by a person.

Images