Comprehensive real-time characterization of ultrasonic signatures from nondestructive evaluation of resistance spot welding process using artificial intelligence

A resistance spot-welding inspection system comprising:

• at least one processor;
• and
• at least one non-transitory computer-readable media storing:
• a machine learning model, that has been trained using a dataset of a plurality of ultrasonic signatures generated during a plurality of resistance spot weld monitoring processes, wherein each of the plurality of ultrasonic signatures includes an M-scan;
• and
• instructions that, when executed by the at least one processor, cause the resistance spot-welding inspection system to perform the following operations:
• a) receiving a sampled ultrasonic signature generated during a monitored resistance spot welding process, wherein the sampled ultrasonic signature includes a sampled M-scan;
• b) using the machine learning model to process the sampled ultrasonic signature;
• c) evaluating the monitored resistance spot-welding process based upon step b);
• and
• d) based upon step c), altering commands for the monitored resistance spot welding process or a subsequent resistance spot welding process.

A resistance spot-welding inspection system comprising:

• at least one ultrasonic sensor generating at least one sampled ultrasonic signature;
• at least one processor;
• and
• at least one non-transitory computer-readable media storing:
• a machine learning model, that has been trained using a dataset of a plurality of ultrasonic signatures generated during a plurality of resistance spot weld monitoring processes, wherein each of the plurality of ultrasonic signatures includes an M-scan, the machine learning model is a neural network including a plurality of convolution layers and a plurality of fully-connected decision layers;
• and
• instructions that, when executed by the at least one processor, cause the resistance spot-welding inspection system to perform the following:
• a) receiving the at least one sampled ultrasonic signature from the at least one ultrasonic sensor generated during a monitored resistance spot welding process in which electrodes are applying current and force based upon commands from a weld controller, wherein the at least one sampled ultrasonic signature includes a sampled M-scan;
• b) using the machine learning model to process the at least one sampled ultrasonic signature;
• and
• c) evaluating the monitored resistance spot-welding process based upon b);
• d) based upon step b) identifying key physical measurements of monitored resistance spot-welding process events in the monitored resistance spot-welding process including at least one event selected from a group consisting of:
• onset of weld nugget, time at which an interface has been penetrated by the weld nugget, rate of weld nugget formation, solidification, and penetration depth into each of a pair of outermost sheets subject to the monitored resistance spot welding process;
• e) altering commands from a weld controller based upon steps b) to d) for a subsequent resistance spot welding process;
• wherein an output of the at least one of the plurality of decision layers encodes a state of the resistance spot-welding process.

A resistance spot-welding inspection system comprising:

• at least one ultrasonic sensor generating at least one sampled ultrasonic signature;
• at least one processor;
• and
• at least one non-transitory computer-readable media storing:
• a machine learning model, that has been trained using a dataset of a plurality of ultrasonic signatures generated during a plurality of resistance spot weld monitoring processes, wherein each of the plurality of ultrasonic signatures includes an M-scan, the machine learning model is a neural network including a plurality of convolution layers and a plurality of fully-connected decision layers;
• and
• instructions that, when executed by the at least one processor, cause the resistance spot-welding system to perform the following:
• a) receiving the at least one sampled ultrasonic signature from the at least one ultrasonic sensor generated during a monitored resistance spot welding process in which electrodes are applying current and force based upon commands from a weld controller, wherein the at least one sampled ultrasonic signature includes a sampled M-scan;
• b) using the machine learning model to process the at least one sampled ultrasonic signature;
• and
• c) evaluating the monitored resistance spot-welding process based upon b);
• d) based upon step b) identifying key physical measurements of monitored resistance spot-welding process events in the monitored resistance spot-welding process including at least one event selected from a group consisting of:
• onset of weld nugget, time at which an interface has been penetrated by the weld nugget, rate of weld nugget formation, solidification, and penetration depth into each of a pair of outermost sheets subject to the monitored resistance spot welding process;
• and
• e) altering commands from a weld controller based upon steps b) to d) for the monitored resistance spot welding process;
• wherein an output of the at least one of the plurality of decision layers encodes a state of the resistance spot-welding process.