Early detection of quality control test failures for manufacturing end-to-end testing optimization

A computing system comprising a computational implementation of a machine-learning-based model (MLM) configured for predicting failure probabilities of product tests scheduled in a future segment of a testing sequence based on test outcomes of different product tests completed during a prior segment of the testing sequence, wherein the computing system is configured for carrying out operations including:

• receiving a set of computer-readable training test-result sequence records (TRSRs), each respective training TRSR comprising a sequence of test results corresponding to a sequence of associated designated product tests (DPTs) previously carried out on a respective unit of a test plurality of units of a particular product,
• wherein each test result of the associated DPT includes a test-value metric (TVM) characterizing a particular aspect of the respective unit tested, and a pass/fail indicator (PFI) indicating whether the respective unit tested passed or failed the associated DPT,
• and wherein the DPTs of the corresponding sequence of associated DPTs are subset of a master set of DPTs for testing the particular product;
• within each of one or more training epochs carried out over the set of training TRSRs, carrying out a respective training iteration over each respective training TRSR of the set, wherein each respective training iteration comprises a sequence of training sub-iterations carried out successively over each given test result of the respective training TRSR, from a first test result to a last test result, wherein each training sub-iteration comprises:
• (i) inputting to the MLM input data comprising the TVM of the given test result and the TVMs of any test results earlier in the sequence than the given test result,
• (ii) based on the input data, computing with the MLM respective predicted probabilities of test failures for the DPTs associated with all test results later in the sequence than the given test result, and
• (iii) based on the PFIs of all the test results later in the sequence than the given test result being applied as ground-truths, training the MLM to increase a statistical confidence of the predicted probabilities;
• and
• upon achieving, from the respective training iterations of the one or more training epochs, one or more thresholds of the statistical confidence of the predicted probabilities, configuring the trained MLM for runtime implementation among one or more computing devices.

A method carried out by a computing system comprising a computational implementation of a machine-learning-based model (MLM), wherein the MLM is configured for predicting failure probabilities of product tests scheduled in a future segment of a testing sequence based on test outcomes of different product tests completed during a prior segment of the testing sequence,

• wherein the MLM is trained with a set of computer-readable training test-result sequence records (TRSRs), each respective training TRSR comprising a sequence of test results corresponding to a sequence of associated designated product tests (DPTs) previously carried out on a respective unit of a test plurality of units of a particular product,
• wherein each test result of the associated DPT includes a test-value metric (TVM) characterizing a particular aspect of the respective unit tested, and a pass/fail indicator (PFI) indicating whether the respective unit tested passed or failed the associated DPT,
• and wherein the DPTs of the corresponding sequence of associated DPTs are a subset of a master set of DPTs for testing the particular product;
• wherein training is arranged to train the MLM to predict, to one or more statistical confidence thresholds, probabilities of test failures for DPTs in a sub-sequence of any given testing sequence, based on input comprising TVMs in the any given testing sequence that precede the sub-sequence,
• and wherein the method comprises:
• inputting to the trained MLM runtime input data comprising runtime TVMs obtained from runtime DPTs of a first segment of a runtime test sequence already carried out on a runtime unit of the particular product, wherein the runtime unit is not part of the test plurality;
• and
• based on the runtime input data, computing with the trained MLM predicted runtime probabilities of runtime test failures for all runtime DPTs of a second segment of the runtime test sequence comprising runtime DPTs not yet carried out on the runtime unit of the particular product, wherein the second segment follows the first segment in the runtime test sequence.