Neural Network Fault Forecasting for Future Machine Operations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning techniques struggle to efficiently predict and classify future faults in complex industrial systems due to the high dimensionality and imbalance of multidimensional time-series data, leading to poor generalization and accuracy in prognostics and diagnostics.
Innovation Solution
A neural network-based system that learns feature representations from historical multidimensional time-series data to embed it into a lower-dimensional space, allowing for accurate prediction and classification of future faults by transforming current data into an embedded space and using future set points to guide predictions, with a specific structure of subnetworks for each prediction step.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning algorithms are applied to predict future faults in complex industrial systems, then fault prediction capability is improved, but the complexity of handling high-dimensional multidimensional time-series data increases
Solution Approach 1:
The patent segments the complex multidimensional time-series data into multiple independent one-dimensional time-series by decomposing multivariate signals. This segmentation reduces the computational complexity of handling high-dimensional data while preserving the essential fault prediction capabilities through individual analysis of each decomposed component.
Solution Approach 2:
The patent introduces wavelet transform as an intermediary technique that bridges the gap between complex multidimensional raw data and the machine learning algorithms. The wavelet transform decomposes the complex data into simpler frequency components, making it more amenable to standard ML prediction models while maintaining prediction accuracy.
2Loss of time
If the prediction horizon is extended to predict faults further in the future, then the useful prediction time is improved, but the prediction accuracy deteriorates
Solution Approach 1:
The patent applies preliminary data preprocessing and feature extraction steps that prepare the data in advance for long-term prediction. By pre-processing the multidimensional time-series data and extracting relevant features before feeding them to the prediction model, the system maintains higher accuracy even when predicting faults further in the future.
3Adaptability or versatility
If multiple faults are considered simultaneously in the prediction model, then the comprehensiveness of fault diagnosis is improved, but the computational burden increases
Solution Approach 1:
The patent segments the multi-fault prediction problem into multiple independent one-dimensional prediction tasks. By decomposing the multidimensional fault signals into separate components and predicting each independently, the system achieves comprehensive multi-fault diagnosis capability while significantly reducing the computational burden compared to modeling all faults simultaneously in a single complex model.
Data Source
AI summary
A system evaluates a plurality of faults in an operation of a machine at a set of future instances of time. The system uses a neural network including a first subnetwork sequentially connected with a sequence of second subnetworks for each of the future instance of time such that an output of one subnetwork is an input to a subsequent subnetwork. The first subnetwork accepts the current time-series data and the current setpoints of operation of the machine. Each of the second subnetworks accepts the output of a preceding subnetwork, an internal state of the preceding subnetwork, and a future setpoint for a corresponding future instance of time. Each of the second subnetworks outputs an individual prediction of each fault of a plurality of faults at the corresponding future instance of time.