PLC Control Program Feature Extraction for Faster ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Industrial automation environments struggle to effectively integrate and train machine learning models to generate control programs, as existing methods are inefficient and time-consuming, and fail to leverage historical data for optimization.
Innovation Solution
A system that integrates a machine learning interface to calculate derivative values, rank and weight them based on their types, and generate feature vectors to train machine learning models for targeted optimization of control programs in industrial automation environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are trained using traditional methods with large amounts of industrial data, then the models can recognize patterns and improve automatically, but the training process requires enormous computing power and time
Solution Approach 1:
The patent extracts only the essential features from industrial data that are relevant to control program optimization, rather than using all available data. The feature extraction component identifies and extracts specific characteristics from control programs and operational data, reducing the data volume required for training while maintaining model accuracy.
Solution Approach 2:
The training process is segmented into multiple stages: feature extraction from control programs, separate training of the machine learning model on extracted features, and deployment. This segmentation allows the model to be trained on a manageable subset of features rather than the entire dataset, reducing computational requirements and training time.
2Adaptability or versatility
If control programs are manually edited in response to operational data, then adjustments can be made to optimize performance, but the process is difficult and time-consuming
Solution Approach 1:
The system enables control programs to self-optimize by automatically generating optimized control programs based on operational data. The machine learning model analyzes operational data and control program performance, then automatically generates optimized control programs without requiring manual intervention from programmers, making the system self-improving.
Solution Approach 2:
The system implements a feedback loop where operational data from industrial processes is continuously collected, analyzed by the machine learning model, and used to generate optimized control programs. This closed-loop feedback mechanism enables automatic adaptation and continuous improvement of control programs based on actual performance data.
3Productivity
If historical data is utilized to generate control programs, then optimization can be achieved, but existing methods do not effectively or efficiently leverage historical data
Solution Approach 1:
The feature extraction component extracts only the relevant historical features from control programs and operational data that are necessary for optimization. Instead of processing all historical data, the system identifies and extracts key features such as control parameters, operational conditions, and performance metrics, reducing processing complexity while maintaining optimization effectiveness.
Solution Approach 2:
The system performs preliminary feature extraction and processing of historical data before the machine learning model training begins. By pre-processing and extracting relevant features from historical control programs and operational data in advance, the system reduces the complexity of real-time data processing and enables more efficient model training and deployment.
Data Source
AI summary
Various embodiments of the present technology generally relate to industrial automation environments. More specifically, embodiments include systems and methods for training a machine learning model for implementation in the environment. In some embodiments, an application receives a data set that comprises a control program configured for implementation by a Programable Logic Controller (PLC). The application processes the data set and calculates derivative values based on the data set. The design application identifies types for individual ones of the feature vectors and ranks the feature vectors based on their types. The design application weights the feature vectors based on their ranks. The design application generates feature vectors that comprise the derivative values and supplies the weighted feature vectors to the machine learning model for training. The application receives a machine learning training output generated by processing the weighted feature vectors.