US 12,248,299 B2
Control system unit for use in a 3-dimensional object manufacturing system and a corresponding method of operating
Gideon May, Amsterdam (NL); Reinder Alexander Lambertus Koelstra, Amsterdam (NL); Yueting Liu, Amsterdam (NL); and Jonathan Van Der Spek, Amsterdam (NL)
Assigned to AM-FLOW HOLDING B.V., Amsterdam (NL)
Appl. No. 17/628,109
Filed by AM-Flow Holding B.V., Amsterdam (NL)
PCT Filed Jul. 16, 2020, PCT No. PCT/NL2020/050467
§ 371(c)(1), (2) Date Jan. 18, 2022,
PCT Pub. No. WO2021/010833, PCT Pub. Date Jan. 21, 2021.
Claims priority of application No. 19186835 (EP), filed on Jul. 17, 2019.
Prior Publication US 2022/0260968 A1, Aug. 18, 2022
Int. Cl. G05B 19/4099 (2006.01)
CPC G05B 19/4099 (2013.01) [G05B 2219/49023 (2013.01)] 5 Claims
OG exemplary drawing
 
1. An operating method for a controller in a 3-dimensional (3D) object production system, the method comprising:
obtaining sensor data signals for a 3D object handled by the 3D object production system;
receiving, by the controller, the sensor data signals for the 3D object handled by the 3D object production system;
classifying, by the controller based on the sensor data signals, the 3D object to a 3D object class of a set of 3D object classes; and
providing, by the controller, a control signal to an actuator unit so as to act on the 3D object in accordance with the 3D object class;
wherein the method further comprises performing a training procedure for each 3D object of a set of one or more 3D objects that at least comprise the 3D object handled by the 3D object production system,
wherein the training procedure comprises, for each 3D object of the set of 3D objects, the following;
receiving object specification data for each 3D object class to be manufactured;
generating, for each 3D object class, synthetic sensor data that is expected to correspond to sensor data of 3D objects manufactured according to object specification data for each 3D object class of the set of 3D object classes; and
training, for each 3D object using the synthetic sensor data and a 3D object class assignment from the set of 3D object classes, the controller in a supervised procedure;
wherein the method comprises providing, by the controller in accordance with the sensor data signals, a feature vector signal representing a computed feature vector,
wherein the classifying the 3D object comprises classifying the 3D object based on a comparison of the computed feature vector with stored labeled reference feature vector data, and
wherein the classifying the 3D object comprises:
comparing the feature vector with the stored labeled reference feature vector data to compute, for each pair of the feature vector and a respective one of a plurality of labeled reference feature vectors of the stored labeled reference feature vector data, a distance vector that is indicative of a distance between the computed feature vector and the respective one of the plurality of labeled reference feature vectors; and
determining a classification score indicating a likelihood that the 3D object to be classified is an object represented by the respective reference feature vector.