Method for automatic quality inspection of an aeronautical part
A method for automatic quality inspection of an aeronautical part, includes detecting faults on an image of the aeronautical part using a trained artificial neural network; training an auto-encoder on a database, by projecting each image of the database onto a small mathematical space in which the images follow a predefined probability law; for each image of the database, calculating a plurality of metrics; supervised training of a classifier from the calculated metrics; detecting faults or anomalies in the image of the aeronautical part using the auto-encoder and the classifier.
1. A method for automatic quality inspection of an aeronautical part, comprising:
training an artificial neural network on a first training database;
acquiring at least one image of the aeronautical part;
detecting faults on the image of the aeronautical part using the trained artificial neural network;
unsupervised training of an auto-encoder on a second training database comprising a plurality of training images of aeronautical parts without fault, comprising the following sub-steps:
projecting each image of the second training database on a mathematical space of smaller dimension than the dimension of each image of the second training database, so that the set of projections of the images of the second database in the mathematical space follows a predefined probability law;
reconstructing each image of the second training database from the projection of the image of the second training database onto the mathematical space to obtain a reconstructed image;
for each image of the second training database, calculating a plurality of metrics from the projection of the image of the second training database onto the mathematical space or from the corresponding reconstructed image, to obtain a metric value for each metric;
supervised training of a classifier from the metric values obtained;
detecting faults or anomalies in the image of the aeronautical part using the trained auto-encoder and classifier;
when the artificial neural network detects at least one fault on the image of the aeronautical or when the artificial neural network detects no fault on the image of the aeronautical part and the classifier detects at least one fault or anomaly in the image of the aeronautical part, the aeronautical part is considered non-compliant
otherwise, the aeronautical part is considered compliant.
2. The method according to claim 1 , wherein the image of the aeronautical part, the plurality of images of the first training database and the plurality of images of the second training database are visible, X-rays, ultrasound or tomography images.
3. The method according to claim 1 , further comprising for each metric, distributing the metric values of the images of the second training database into a first group of values and a second group of values according to a predefined validity criterion, the classifier being trained to consider the images of the second training database having the metric values of the first group of values as having no faults or anomalies and the images of the second training database having the metric values of the second group of values as images to be rejected.
4. The method according to claim 1 , wherein detecting a fault on the image of the aeronautical part using the trained auto-encoder and classifier comprises the following sub-steps:
projecting, by the auto-encoder, of the image of the aeronautical part onto the mathematical space;
reconstructing, by the auto-encoder, of the image of the aeronautical part from the projection of the image of the aeronautical part onto the mathematical space to obtain a reconstructed image of the aeronautical part
calculating metrics from the projection of the image of the aeronautical part onto the mathematical space or from the reconstructed image of the aeronautical part to obtain metric values;
detecting faults on the image of the aeronautical part by the classifier from the metric values obtained for the image of the aeronautical part.
5. The method according to claim 1 , wherein the auto-encoder is of variational type or adversarial-auto-encoder type.
6. The method according to claim 1 , wherein the plurality of metrics comprises, for a given image, a distance between the projection of the image on the mathematical space and the set of the projections of the plurality of images of the second database in the mathematical space, and/or a distance between the image and the reconstructed image, and/or an entropy of the gaps between the image and the reconstructed image.
7. The method according to claim 6 , wherein a law of probability is a multivariate Gaussian law and the distance between the projection of the image onto the mathematical space and the set of the projections of the plurality of images of the second database onto the mathematical space is the Mahalanobis distance.
8. The method according to claim 1 , further comprising generating an alert when the artificial neural network detects no fault on the image of the aeronautical part and the classifier detects at least one fault or anomaly in the image of the aeronautical part.
9. A non-transitory computer readable instructions comprising program code instructions for the execution of the steps of the method according to claim 1 when said program code instructions are executed on a computer.
10. The method according to claim 1 , wherein the training of the artificial neural network is a supervised training and the first training database comprises a plurality of training images of aeronautical parts with or without fault.
11. The method according to claim 1 , wherein the training of the artificial neural network is an unsupervised training and the first training database comprises a plurality of training images of aeronautical parts without fault.