Valve Vibration Fault Detection Using LSTM-AE Signal Indicators
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing valve fault detection methods for reciprocating pumps, particularly plunger pumps, face challenges in accuracy due to insufficient consideration of time-domain and frequency-domain characteristics, lack of comprehensive indicator systems, reliance on single neural network models, limited real-world data, and absence of expert experience integration, leading to low prediction precision and applicability in industrial settings.
Innovation Solution
A valve fault detection method that constructs a comprehensive signal characteristic indicator system combining time-domain and frequency-domain analysis, utilizes an LSTM-AE model for pattern recognition, and integrates expert experience through a structured library to enhance prediction precision and universality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single neural network model is used for fault detection, then the device complexity is low, but the prediction precision is insufficient
Solution Approach 1:
The patent combines multiple neural network models (LSTM, CNN, and Autoencoder) into an integrated fault detection system. Each model processes different aspects of the vibration signal characteristics, and their outputs are fused to generate the final fault diagnosis result, thereby improving prediction precision through ensemble learning while managing complexity through modular architecture design
Solution Approach 2:
The patent segments the fault detection task into multiple specialized sub-tasks handled by different neural network components. The LSTM model handles temporal sequence modeling, the CNN model extracts spatial features from signal representations, and the Autoencoder performs anomaly detection. This segmentation allows each component to specialize in specific feature extraction, improving overall prediction precision
2Measurement precision
If only time-domain characteristics are considered in the indicator system, then the system complexity is low, but the measurement precision of fault detection is insufficient
Solution Approach 1:
The patent transitions from single-dimensional time-domain analysis to multi-dimensional analysis by incorporating both time-domain characteristics (mean, variance, skewness, kurtosis) and frequency-domain characteristics (spectral centroid, bandwidth, peak frequencies). This dimensional expansion enables the indicator system to capture fault information from multiple signal aspects, significantly improving fault detection accuracy
Solution Approach 2:
The patent creates a composite indicator system that integrates multiple types of signal characteristics (time-domain statistics, frequency-domain features, and time-frequency representations) into a unified feature set. This composite approach combines the strengths of different analysis domains, enabling more comprehensive fault detection while maintaining systematic organization of the indicators
3Adaptability or versatility
If detection is based on laboratory environment data only, then the model training is simple, but the adaptability to real industrial operations is poor
Solution Approach 1:
The patent performs preliminary data collection and annotation during the laboratory phase, creating a comprehensive dataset that includes various fault conditions, operating parameters, and environmental variations. This preliminary action ensures that the model is exposed to diverse scenarios during training, improving its adaptability to real industrial operations without requiring complex on-site data collection later
Solution Approach 2:
The patent designs the indicator system and neural network models to be universally applicable across different pump types and operating conditions. By training on diverse laboratory data covering multiple fault modes and operational scenarios, the model learns generalized fault patterns that can be transferred to real industrial settings, enhancing model universality while avoiding overfitting to specific equipment
Data Source
AI summary
A valve fault detection method includes: constructing a valve fault indicator system reflecting characteristics of vibration signals of valve states; establishing an LSTM-AE model by using an LSTM deep model and according to an autoencoder principle, and setting a fault discrimination threshold; inputting training data into the LSTM-AE model based on the valve fault indicator system, to train the LSTM-AE model; and inputting detection data of a to-be-detected valve into the trained LSTM-AE model based on the valve fault indicator system, and comparing an output value obtained by the LSTM-AE model with the set fault discrimination threshold, to judge whether the detected valve is faulty and outputting a first judgment result.