Machine Learning Controller Safety Circuits for Noise-Robust Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In manufacturing environments, machine learning devices in controllers for manufacturing machines are prone to producing abnormal outputs due to noise in input data, leading to inaccurate learning and inference results, which can result in machining errors and tool damage.
Innovation Solution
A controller with a machine learning device that includes a state observation unit, an input safety circuit for detecting and correcting abnormal input data, a machine learning unit for learning and inference using safe data, and an output safety circuit for detecting and correcting abnormal inference data, ensuring the output of safe and accurate results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning device performs learning and inference using input data from manufacturing machine, then learning accuracy and inference reliability are improved, but abnormal values such as noise in factory environment cause abnormal output and machining errors
Solution Approach 1:
The patent applies preliminary action by performing safety checks on input data before it is used for learning and inference. The safety circuit detects abnormal values in advance and prevents them from entering the machine learning processing pipeline, thereby protecting the system from noise and ensuring reliable inference results.
Solution Approach 2:
The patent introduces a safety circuit as an intermediary between the input data acquisition and the machine learning unit. This intermediary component detects and filters abnormal values in input data, acting as a mediator that protects the learning system from harmful noise while allowing valid data to pass through for accurate learning and inference.
2Manufacturing precision
If machine learning device is used for machine control with internal controller data, then control precision is improved, but noise in input data causes abnormal inference output
Solution Approach 1:
The patent applies preliminary anti-action by implementing a safety circuit that proactively detects and counteracts the harmful effects of noise in input data before it can affect the machine learning process. The safety circuit identifies abnormal values and prevents them from influencing the learning model or inference results, thereby maintaining control precision despite noisy factory environment data.
3Adaptability or versatility
If parameter is probabilistically determined in machine learning, then learning flexibility is improved, but abnormal inference data may be output even with normal input data
Solution Approach 1:
The patent applies feedback by implementing a safety circuit that monitors inference output and provides feedback when abnormal values are detected. Even though the machine learning unit operates with probabilistic parameters for flexible learning, the safety circuit continuously checks the output and can trigger re-inference or error handling when abnormal results occur, thereby maintaining inference reliability alongside learning flexibility.
Data Source
AI summary
In a controller and a machine learning device capable of suppressing an influence of an abnormal value based on noise, etc., the machine learning device included in the controller includes a state observation unit for acquiring input data including at least one of internal data and external data of the manufacturing machine controlled by the controller, an input safety circuit for detecting an abnormality in the input data and outputting safe input data, a machine learning unit for executing learning of a learning model and inference using the learning model based on the safe input data and outputting inference data as an inference result, an output safety circuit for detecting an abnormality in the inference data and outputting safe inference data, and an output unit for outputting output data based on the safe inference data.