Systems and methods for identifying manufacturing defects

A system for classifying manufacturing defects, the system comprising:

• processor;
• and
• memory, wherein the memory has stored therein instructions that, when executed by the processor, cause the processor to:
• identify, from a training dataset, a first data sample and a second data sample;
• identify a first label predicted for the first data sample;
• compare the first label to a second label assigned to the first data sample;
• determine a labeling error for the first label based on the comparing, wherein the processor being configured to determine the labeling error include the processor being configured to estimate a joint probability distribution between the first label and the second label;
• based determining the labeling error for the first label:
• remove, from the training dataset, the first data sample and outputting a filtered training dataset including the second data sample;
• train a first machine learning model with the filtered training dataset;
• and
• train a second machine learning model based on at least one of the first data sample or the second data sample;
• receive a second dataset including first product data and second product data associated with a manufactured product, wherein the first product data includes a first image of a product in a production line and the second product data includes a second image of the product in the production line;
• invoke the second machine learning model for predicting a first confidence of the first product data and a second confidence of the second product data;
• remove the second product data from the second dataset based on determining the second confidence and outputting a modified second dataset;
• invoke the first machine learning model for generating a classification based on the modified second dataset;
• detect, based on the classification, that the first image contains a manufacturing defect of the product in the production line;
• and
• remove the product from the production line based on detecting the manufacturing defect.

A system for classifying manufacturing defects, the system comprising:

• a data collection circuit configured to collect an input dataset;
• and
• a processing circuit coupled to the data collection circuit, the processing circuit having logic for:
• identifying, from a training dataset, a first data sample and a second data sample;
• identifying a first label predicted for the first data sample;
• comparing the first label to a second label assigned to the first data sample;
• determining a labeling error for the first label based on the comparing, wherein the determining of the labeling error is based on estimating a joint probability distribution between the first label and the second label;
• based on determining the labeling error for the first label:
• removing, from the training dataset, the first data sample and outputting a filtered training dataset including the second data sample;
• training a first machine learning model with the filtered training dataset;
• and
• training a second machine learning model based on at least one of the first data sample or the second data sample;
• receiving a second dataset including first product data and second product data associated with a manufactured product, wherein the first product data includes a first image of a product in a production line and the second product data includes a second image of the product in the production line;
• invoking the second machine learning model for predicting a first confidence of the first product data and a second confidence of the second product data;
• removing the second product data from the second dataset based on determining the second confidence and outputting a modified second dataset;
• invoking the first machine learning model for generating a classification based on the modified second dataset;
• detecting, based on the classification, that the first image contains a manufacturing defect of the product in the production line;
• and
• removing the product from the production line based on detecting the manufacturing defect.