Agricultural Machine Anomaly Control Using Normal-State Autoencoders
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current control systems for agricultural machines are unable to detect unforeseen anomalies, as they require pre-defined training data to recognize normal and abnormal states, limiting their ability to respond to unknown anomalies effectively.
Innovation Solution
A control system that learns a model of the normal state using training data and compares current operating variables to recognize anomalous states, utilizing an autoencoder neural network to compress and decompress data, allowing for detection of rare operating conditions without prior knowledge of anomalies, and continuously trains using operational data for improved selectivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If supervised learning with pre-defined training data is used to detect anomalies, then the system can reliably distinguish normal from known abnormal states, but it cannot detect unforeseen anomalies that were not included in the training data
Solution Approach 1:
Instead of training the system to recognize abnormal states (supervised learning), the patent inverts the approach by training the system to recognize only normal states. Any deviation from the learned normal state is automatically flagged as an anomaly, enabling detection of previously unseen abnormal conditions without requiring training examples of those specific anomalies.
Solution Approach 2:
The system performs self-training by automatically learning the normal operating state from training data without requiring manual definition of anomaly types. The autoencoder neural network self-optimizes to compress and reconstruct normal operating patterns, enabling the system to autonomously adapt to the machine's normal behavior and detect deviations.
2Reliability
If traditional control systems with pre-defined anomaly ranges are used, then known anomalies can be detected, but the system complexity increases exponentially with the number of operational variables
Solution Approach 1:
The patent replaces traditional mechanical/statistical anomaly detection methods (comparing values against pre-defined ranges) with a neural network-based autoencoder system. This substitution enables the system to handle high-dimensional operational variables (including images) without exponential complexity growth, as the neural network learns complex patterns automatically rather than requiring exhaustive rule definition.
Solution Approach 2:
The system changes the parameter representation by using compression ratios and reconstruction errors as anomaly indicators instead of direct comparison with pre-defined thresholds. This parameter transformation allows the system to detect anomalies across multiple operational variables simultaneously without requiring exponential increases in system complexity.
3Adaptability or versatility
If autoencoder neural networks are used to compress and decompress operating data, then unforeseen anomalies can be detected through reconstruction error, but training data requirements increase
Solution Approach 1:
The system performs preliminary training during a dedicated training phase using historical normal operating data before deployment. This preliminary action allows the autoencoder to learn and compress the normal operating patterns in advance, so that during actual operation, the system can efficiently detect anomalies without requiring additional training data collection.
Data Source
Figure 1
Figure 2
AI summary
A control system (13) for an agricultural machine (1) has at least one data input (14) for receiving a current set of operating parameters of the machine (1) that is characteristic of a current operating state of the machine (1) and a control output (14) for issuing a control command that influences the operation of the machine (1). The control system (13) has a model (24) of the normal state of the machine, learned from training data, and is configured to detect a current anomalous state of the machine (1) by comparing the current set of operating parameters with the model (24) and to react to it by issuing the control command.