Control Signal Anomaly Detection Using Multi-Decoder Autoencoders
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly recognition methods in control systems lack robustness in extrapolation and often require extensive training data, limiting their effectiveness in real-world applications.
Innovation Solution
The method employs a variational autoencoder with multiple decoders to generate diverse hypotheses for reconstructing input signals, allowing for improved anomaly recognition and control signal generation. This approach includes a training system using a discriminator to optimize the autoencoder's parameters, ensuring robustness and reduced training data requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing anomaly recognition methods are used, then the system can detect anomalies, but the robustness in extrapolation is poor and extensive training data is required
Solution Approach 1:
The patent segments the anomaly detection task into multiple hypotheses generated by different decoders, each handling specific aspects of signal reconstruction. This segmentation allows the system to evaluate multiple potential explanations for input signals, improving robustness without requiring proportionally more training data
Solution Approach 2:
The patent introduces a discriminator as an intermediary component that mediates between the autoencoder hypotheses and the final anomaly detection decision. The discriminator evaluates the plausibility of each hypothesis, enabling robust extrapolation by learning from limited training data how to assess hypothesis quality without needing extensive annotated anomaly data
2Adaptability or versatility
If multiple hypotheses are generated for signal reconstruction, then the capacity for extrapolation is improved, but the device complexity increases
Solution Approach 1:
The patent implements multiple decoders that share common encoder components and intermediate representations, allowing the system to generate multiple hypotheses while reusing computational resources. This multi-functionality approach enables improved extrapolation capacity without proportionally increasing overall device complexity
Solution Approach 2:
The patent creates simplified copies of the decoding process through multiple decoders that operate on the same encoded representations. Rather than building entirely separate complex systems for each hypothesis, the patent uses copying of the decoding architecture with different parameter configurations, reducing the complexity burden of generating multiple hypotheses
Data Source
AI summary
A computer-implemented method for classifying an input signal that is ascertained as a function of an output signal of a sensor as to whether or not it has an anomaly, it being decided, as a function of an output signal of an autoencoder (60) to which the input signal is supplied, whether or not the input signal has the anomaly. The autoencoder has at least one encoder and at least one decoder. An intermediate variable is ascertained as a function of the input signal by the encoder, and the output signal is ascertained by the decoder as a function of the intermediate variable. The autoencoder provides, for the input signal supplied to it, a plurality of hypotheses of formula for the reconstruction of the input signal, and the output signal is ascertained as a function of this plurality of hypotheses of formula.