LSTM Anomaly Detection for Multi-Sensor Substrate Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In manufacturing processes, particularly in semiconductor and display industries, identifying anomalies and their causes from sensor data is laborious, time-consuming, and requires domain expertise, making it difficult to scale and automate the detection process.
Innovation Solution
A processing device trains a Long Short-Term Memory (LSTM) recurrent neural network (RNN) model using sensor data from multiple sensors to analyze and detect anomalies, reducing the need for manual trial and error by using reconstruction data for semiconductor processing fault detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual root cause analysis is conducted at the time of failure or anomaly occurrence, then domain expertise can be applied to identify issues, but the process becomes time-consuming and causes downtime
Solution Approach 1:
The system performs preliminary anomaly detection and root cause analysis using the trained LSTM model before actual failure occurs. By continuously monitoring sensor data and comparing it against the trained model's expectations, the system identifies anomalies early and determines their causes in advance, eliminating the need for time-consuming manual analysis at the moment of failure and thus reducing downtime while maintaining detection accuracy
2Measurement precision
If manual trial and error methods are used for anomaly detection, then domain expertise can guide the process, but the detection becomes laborious and difficult to scale
Solution Approach 1:
The patent replaces manual mechanical trial-and-error analysis with an automated LSTM recurrent neural network model. The model is trained on historical sensor data to learn normal process patterns, then automatically detects anomalies and identifies their causes by comparing real-time sensor data against learned patterns. This substitution eliminates laborious manual intervention while maintaining or improving detection accuracy, and enables scalable deployment across multiple manufacturing processes and equipment
3Loss of information
If complex multi-sensor data is analyzed manually, then comprehensive root cause analysis can be performed, but the processor overhead and energy consumption increase
Solution Approach 1:
The system performs preliminary processing by training the LSTM model on historical multi-sensor data to capture normal process patterns and relationships. During operation, the trained model efficiently processes incoming sensor data by comparing it against learned patterns, automatically identifying anomalies and their causes. This preliminary training phase consolidates the complex analysis work, enabling fast, low-overhead processing during actual operation while maintaining complete root cause analysis capability
Solution Approach 2:
The LSTM model creates a virtual copy or representation of normal manufacturing process patterns by training on historical data. This learned model serves as a reference that can be quickly compared against real-time sensor data without requiring complex manual analysis of the original multi-sensor streams. The copying approach reduces processor overhead during operation while preserving the ability to perform comprehensive root cause analysis through the trained model's predictions
Data Source
AI summary
A method includes identifying current sensor data associated with processing of substrates by substrate processing equipment. The method further includes providing the current sensor data as input to a trained machine learning model. The trained machine learning model is trained using training input comprising a first window of time of historical sensor data and target output comprising the first window of time or a second window of time of the historical sensor data to generate the trained machine learning model. The historical sensor data is associated with normal runs of processing of historical substrates by the substrate processing equipment. The method further includes obtaining, from the trained machine learning model, one or more outputs. The method further includes causing, based on the one or more outputs, an anomaly response action associated with the substrate processing equipment.