Artificial intelligence anomaly detection using x-ray computed tomography scan data
Provided herein are methods, apparatuses, computer program products, and systems for anomaly detection using machine learning models. One method can include obtaining X-ray computed tomography (CT) scan data for a scan object of a predetermined object type; producing derived data from the X-ray CT scan data, wherein the derived data reveals at least one interior structure usable for anomaly detection in objects of the predetermined object type; inputting at least the derived data into a machine learning model, which has been trained using at least derived data produced from prior X-ray CT scan data for objects of the predetermined object type; receiving an output from the machine learning model that indicates that at least one anomaly has been detected for the scan object; and providing an output to a physical device based on the at least one anomaly having been detected for the scan object.
1 . A system comprising:
a data processing apparatus including at least one processor; and
a non-transitory computer-readable medium encoding instructions configured to cause the data processing apparatus to perform operations comprising:
obtaining X-ray computed tomography (CT) scan data for a scan object of a predetermined object type,
producing derived data comprising multiple views of the scan object generated from the X-ray CT scan data, wherein the derived data reveals at least one interior structure usable for anomaly detection in objects of the predetermined object type, wherein each of the multiple views is from a different location around the scan object,
inputting at least the derived data into a machine learning model, which has been trained using at least derived data comprising multiple views of objects produced from prior X-ray CT scan data for the objects of the predetermined object type,
receiving an output from the machine learning model that indicates that at least one anomaly has been detected for the scan object, and
providing an output to a physical device based on the at least one anomaly having been detected for the scan object, wherein
the derived data further comprises: (i) a shape representation of the X-ray CT scan data and (ii) feature data of one or more voids or contaminants detected in the X-ray CT scan data,
the shape representation reveals at least one interior structure usable for anomaly detection in objects of the predetermined object type,
the feature data of the one or more voids or contaminants reveals at least one interior structure usable for anomaly detection in objects of the predetermined object type, and
the machine learning model has been trained using at least the derived data comprising (i) the multiple views of objects, (ii) shape representations, and (iii) feature data of voids or contaminants produced from the prior X-ray CT scan data for the objects of the predetermined object type.
2 . The system of claim 1 , wherein the shape representation represents information about material properties of the scan object.
3 . The system of claim 1 , wherein the producing comprises generating the shape representation using a segmentation method that is based on a structural property of the X-ray CT scan data.
4 . The system of claim 1 , wherein the shape representation comprises a series of segmentations of the X-ray CT scan data, wherein the series of segmentations are generated using multiple threshold values that segment different materials from the X-ray CT scan data.
5 . The system of claim 1 , wherein the derived data comprises dimensional deviation data on a primitive-matched feature, wherein the primitive-matched feature is a geometric shape with primitive fit to the X-ray CT scan data.
6 . The system of claim 5 , wherein the machine learning model generates the output that indicates the at least one anomaly by processing the dimensional deviation data on the primitive-match feature using a quality criterion that describes allowable variation of the dimensional deviation data.
7 . The system of claim 5 , wherein the dimensional deviation data on the primitive-matched feature comprises concentricity, alignment, and spacing of the primitive-matched feature.
8 . The system of claim 5 , wherein the producing comprises:
determining a deviation type of the primitive-matched feature based on the predetermined object type of the scan object; and
generating, based on the deviation type of the primitive-matched feature, the dimensional deviation data on the primitive-matched feature.
9 . The system of claim 5 , wherein the dimensional deviation data on the primitive-matched feature comprises a plane-to-plane distance.
10 . The system of claim 1 , wherein the providing comprises detecting the scan object as a counterfeit object based on the at least one anomaly having been detected for the scan object.
11 . The system of claim 1 , wherein the providing comprises:
detecting a change to a production machine that processes the scan object on a product handling line based on the at least one anomaly having been detected for the scan object; and
outputting the detected change to the production machine.
12 . The system of claim 11 , wherein detecting the change to the production machine comprises detecting wear and tear of a tool of the production machine.
13 . The system of claim 1 , wherein the X-ray CT scan data comprises three-dimensional (3D) reconstruction data for the scan object, and the multiple views are rendered views of the 3D reconstruction data produced using a rendering setting that accentuates material differences.
14 . The system of claim 1 , wherein each different location around the scan object is specified by a vertex of a polyhedron.
15 . The system of claim 1 , wherein the scan object is an assembly comprising two or more parts, and the providing comprises performing assembly verification based on the at least one anomaly having been detected for the scan object.
16 . The system of claim 1 , wherein the providing comprises:
detecting a change in a supply chain based on the at least one anomaly having been detected for the scan object; and
outputting the detected change in the supply chain.
17 . The system of claim 1 , wherein the providing comprises: processing the at least one anomaly using a classifier to determine an anomaly type from a predetermined number of anomaly types for the at least one anomaly.
18 . The system of claim 1 , wherein the providing comprises:
detecting a variance in production lines processing a same type of products; and
outputting the detected variance in the production lines.
19 . A system, comprising:
a data processing apparatus including at least one processor; and
a non-transitory computer-readable medium encoding instructions configured to cause the data processing apparatus to perform operations comprising:
obtaining X-ray computed tomography (CT) scan data for a scan object of a predetermined object type,
producing derived data comprising multiple views of the scan object generated from the X-ray CT scan data, wherein the derived data reveals at least one interior structure usable for anomaly detection in objects of the predetermined object type, wherein each of the multiple views is from a different location around the scan object,
inputting at least the derived data into a machine learning model, which has been trained using at least derived data comprising multiple views of objects produced from prior X-ray CT scan data for the objects of the predetermined object type,
receiving an output from the machine learning model that indicates that at least one anomaly has been detected for the scan object,
providing an output to a physical device based on the at least one anomaly having been detected for the scan object,
obtaining a training data set comprising at least the derived data produced from the prior X-ray CT scan data for training objects of the predetermined object type, wherein the derived data produced from the prior X-ray CT scan data reveals at least one interior structure usable for anomaly detection in objects of the predetermined object type,
training the machine learning model on the training data set to form a trained machine learning model, and
determining an operation parameter of the trained machine learning model based on a throughput requirement of a product handling process and a tolerance for incorrectly predicted objects of the product handling process.
20 . The system of claim 19 , wherein the X-ray CT scan data comprises three-dimensional (3D) reconstruction data for the scan object, and the multiple views are rendered views of the 3D reconstruction data produced using a rendering setting that accentuates material differences.
21 . The system of claim 19 , wherein each different location around the scan object is specified by a vertex of a polyhedron.
22 . The system of claim 19 , wherein the providing comprises: processing the at least one anomaly using a classifier to determine an anomaly type from a predetermined number of anomaly types for the at least one anomaly.
23 . A non-transitory computer-readable medium encoding instructions operable to cause data processing apparatus to perform operations comprising:
obtaining X-ray computed tomography (CT) scan data for a scan object of a predetermined object type;
producing derived data comprising feature data of one or more voids or contaminants detected in the X-ray CT scan data, wherein the feature data of the one or more voids or contaminants reveals at least one interior structure usable for anomaly detection in objects of the predetermined object type;
inputting at least the derived data into a machine learning model, which has been trained using at least derived data comprising feature data of voids or contaminants detected in prior X-ray CT scan data for objects of the predetermined object type;
receiving an output from the machine learning model that indicates that at least one anomaly has been detected for the scan object; and
providing an output to a physical device based on the at least one anomaly having been detected for the scan object, wherein the output to the physical device causes a change in an upstream process or a change in a vendor in a supply chain determined based on the at least one anomaly,
wherein the one or more voids or contaminants comprise one or more inclusions detected in the X-ray CT scan data, the feature data comprising inclusion data of the one or more inclusions, wherein the scan object was manufactured using an injection molding process, and the output to the physical device comprises data characterizing agglomerations in the injection molding process of fiber-filled resins.
24 . The non-transitory computer-readable medium of claim 23 , wherein the one or more voids or contaminants comprise one or more pores detected in the X-ray CT scan data, the feature data comprising porosity data of the one or more pores, wherein the scan object was manufactured by molding, bonding, or sealing.
25 . The non-transitory computer-readable medium of claim 23 , wherein the feature data comprises location data of at least one of the one or more voids or contaminants.
26 . The non-transitory computer-readable medium of claim 23 , wherein the feature data comprises size data of at least one of the one or more voids or contaminants.
27 . The non-transitory computer-readable medium of claim 23 , wherein the feature data comprises shape data of at least one of the one or more voids or contaminants.
28 . The non-transitory computer-readable medium of claim 23 , wherein the feature data comprises distance to surface data of at least one of the one or more voids or contaminants.
29 . The non-transitory computer-readable medium of claim 23 , wherein the feature data comprises a count of the one or more voids or contaminants.
30 . The non-transitory computer-readable medium of claim 23 , wherein the feature data comprises sphericity data of at least one of the one or more voids or contaminants.