System and method for generating machine learning model with trace data

1. A method for classifying an electronic device manufactured via one or more manufacturing processes, the method comprising:

receiving a plurality of time-series sensor data obtained in the one or more manufacturing processes of the electronic device, wherein the time-series sensor data has a first size and a first ordering;

arranging the plurality of time-series sensor data in a two-dimensional (2D) data array, wherein the arranging includes:

processing the time-series sensor data to have at least one of a second size different from the first size, or to have a second ordering different from the first ordering; and

storing the time-series sensor data in the 2D data array according to at least the second size or the second ordering;

providing the 2D data array to a convolutional neural network model;

identifying, using the convolutional neural network model, a pattern in the 2D data array that correlates to a fault condition;

classifying, using the convolutional neural network model, whether the electronic device includes a fault based on the pattern correlating to the fault condition;

generating an output for addressing the fault in the electronic device; and

adjusting an element of the one or more manufacturing processes based on the output.

2. The method of claim 1 , wherein the electronic device is a display device.

3. The method of claim 1 , wherein the electronic device is a glass panel of a display device.

4. The method of claim 1 , wherein the convolutional neural network model is VGG, Resnet, Alexnet, or Squeezenet.

5. The method of claim 1 , wherein a result of the classifying includes a type and a location of the fault condition.

6. The method of claim 1 , wherein the plurality of time-series sensor data is separated from each other within the 2D data array.

7. The method of claim 1 , further comprising:

discarding the electronic device that includes the fault during the one or more manufacturing processes.

8. The method of claim 1 , further comprising:

placing the plurality of time-series sensor data into a raw trace data array having a first dimension corresponding to a number of sensors and a second dimension corresponding to a number of data samples for each of the sensors;

determining whether the raw trace data array fits into the 2D data array;

splitting the raw trace data array into a plurality of raw trace data arrays; and

placing the plurality of raw trace data arrays into the 2D data array.

9. The method of claim 8 , wherein each of the plurality of raw trace data arrays includes an overlap area.

10. The method of claim 1 , further comprising:

receiving a plurality of time-series sensor data sets having different numbers of data samples;

determining a time-series sensor data set among the plurality of time-series sensor data sets that has a first number of data samples; and

resizing remaining sensor time-series data sets of the plurality of time-series sensor data sets to have the first number of data samples.

11. The method of claim 1 , further comprising:

determining a subset of the plurality of time-series sensor data that correlate to the fault condition; and

rearranging the 2D data array to place the subset of plurality of time-series sensor data in a vicinity to each other.

12. The method of claim 1 , further comprising:

receiving a plurality of time-series sensor data sets, wherein each of the plurality of time-series sensor data sets includes data corresponding to a plurality of processing steps in the one or more manufacturing processes;

comparing data sizes of each of the plurality of processing steps in the plurality of time-series sensor data sets;

for each processing step of the plurality of processing steps, determining a time-series sensor data set among the plurality of time-series sensor data sets that has a first step size;

mapping time-series sensor data corresponding to a respective process step in remaining time-series sensor data sets to have the first step size; and

generating the 2D data array for each of the plurality of time-series sensor data sets by combining respective process steps that are mapped to the first step size.

13. The method of claim 12 , further comprising:

selecting first data samples of the first step size when mapping the time-series sensor data.

14. The method of claim 12 , further comprising:

resampling the time-series sensor data with decimation to obtain a resampled time-series sensor data having the first step size.

15. The method of claim 1 , further comprising:

determining an average of the plurality of time-series sensor data;

shifting each of the plurality of time-series sensor data to place the average at a zero; and

normalizing each of the plurality of time-series sensor data within a range between −1 and +1.

16. The method of claim 1 , wherein the one or more manufacturing processes are selected from deposition, removal, patterning, and a combination thereof.

17. A method for classifying an electronic device manufactured via one or more manufacturing processes, the method comprising:

providing a convolutional neural network model that is pre-trained with a first set of data having a first format;

receiving a second set of data having a second format, wherein the second set of data has a first size and a first ordering;

converting the second format of the second set of data to the first format and generating a third set of data;

re-training the convolutional neural network model with the third set of data;

obtaining an updated convolutional neural network model;

generating, using the updated convolutional neural network model, an output for addressing a fault in the electronic device; and

adjusting an element of the one or more manufacturing processes based on the output,

wherein the converting includes:

processing the second set of data to have at least one of a second size different from the first size, or to have a second ordering different from the first ordering; and

storing the second set of data in a data array according to at least the second size or the second ordering.

18. The method of claim 17 , wherein the second set of data is a plurality of time-series sensor data obtained in the one or more manufacturing processes, and the first format is a two-dimensional (2D) data array, and the method further comprising:

providing the third set of data having the first format of the 2D data array to the updated convolutional neural network model;

identifying, using the updated convolutional neural network model, a pattern in the 2D data array that correlates to a fault condition;

classifying, using the updated convolutional neural network model, whether the electronic device includes the fault based on the pattern correlating to the fault condition; and

at least one of:

1) Placing the plurality of time-series sensor data into a raw trace data array having a first dimension corresponding to a number of sensors and a second dimension corresponding to a number of data samples for each of the sensors;

determining whether the raw trace data array fits into the 2D data array;

splitting the raw trace data array into a plurality of raw trace data arrays; and

placing the plurality of raw trace data arrays into the 2D data array;

2) Receiving a plurality of time-series sensor data sets having different numbers of data samples;

determining a time-series sensor data set among the plurality of time-series sensor data sets that has a first number of data samples; and

resizing remaining sensor time-series data sets of the plurality of time-series sensor data sets to have the first number of data samples;

3) Determining a subset of the plurality of time-series sensor data that correlate to the fault condition; and

rearranging the 2D data array to place the subset of plurality of time-series sensor data in a vicinity to each other; or

4) Receiving a plurality of time-series sensor data sets, wherein each of the plurality of time-series sensor data sets includes data corresponding to a plurality of processing steps in the one or more manufacturing processes;

comparing data sizes of each of the plurality of processing steps in the plurality of time-series sensor data sets;

for each processing step of the plurality of processing steps, determining a time-series sensor data set among the plurality of time-series sensor data sets that has a first step size;

mapping time-series sensor data corresponding to a respective process step in remaining time-series sensor data sets to have the first step size; and

generating the 2D data array for each of the plurality of time-series sensor data sets by combining respective process steps that are mapped to the first step size.

19. The method of claim 18 , wherein the electronic device is a glass panel of a display device.

20. The method of claim 18 , wherein the updated convolutional neural network model is VGG, Resnet, Alexnet, or Squeezenet.

21. The method of claim 18 , further comprising:

discarding the electronic device that includes the fault during the one or more manufacturing processes.