Rotating Machinery Condition Assessment via Neural Network Motor Current Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for condition monitoring of rotating machinery connected to electric motors are invasive, require shutdown for sensor installation, and face challenges in setting alarm thresholds due to transmission path effects and complex interactions between rotor and stator fluxes, especially when the electrical machine is supplied by a drive.
Innovation Solution
A method using neural networks to analyze motor currents and voltages, eliminating the need for electromechanical sensors like accelerometers, by acquiring and processing electrical signals to assess the health of rotating machinery through feature extraction and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If oil debris sensors or vibration sensors are mounted on drive train components, then condition monitoring capability is improved, but the operation of the machine must be stopped for sensor installation
Solution Approach 1:
The patent replaces mechanical sensors (oil debris sensors, vibration sensors) with electrical signal analysis. By analyzing electrical signals already present in the motor system, the invention eliminates the need for physical sensor installation on drive train components, thus avoiding downtime while maintaining condition monitoring capability
Solution Approach 2:
The patent uses electrical signals as an intermediary to indirectly monitor the condition of rotating machinery. Instead of directly measuring mechanical parameters with physical sensors, the system uses electrical current signals that reflect the mechanical state, enabling non-invasive monitoring
2Ease of operation
If casing-mounted accelerometers are used for non-invasive monitoring, then no shutdown is required, but transmission path effects and installation variations make it difficult to set alarm threshold levels
Solution Approach 1:
The patent extracts the monitoring function from the mechanical transmission path by analyzing electrical signals at the motor level. This removes the influence of transmission path effects and installation variations that affect casing-mounted accelerometers, as the electrical signals are generated at the source before mechanical transmission occurs
Solution Approach 2:
Instead of monitoring mechanical vibrations and working backwards to infer component health, the patent inverts the approach by monitoring electrical signals and using them to directly assess the condition of rotating machinery, bypassing the mechanical transmission path entirely
3Ease of operation
If electrical signals are analyzed to evaluate rotating machinery health, then invasive sensors are eliminated, but the complex interactions between rotor and stator fluxes and inverter control actions distort the voltage supplied to the electric motor
Solution Approach 1:
The patent applies preliminary signal processing and neural network training to account for the complex motor-inverter interactions. By pre-training the neural network with data that includes these distortions, the system learns to recognize fault patterns despite the signal complexity, effectively handling the distorted voltage signals
Data Source
AI summary
The subject of the invention is a method and a computer program for assessing the condition of rotating machinery connected to an electric motor, where the motor (MM) and the rotating machine (RM) comprise a drive train electrically powered from a three phase motor and said motor is connected with a computing device. The invention is characterized in that is comprises the step of classifying the condition of rotating machine (RM) connected to the electric motor (MM) of a drive train with unknown condition, by analyzing all the extracted representative features obtained from the electric signals acquired from the electric motor (MM) of said drive train, by using a trained neural network whose twining was conducted by analyzing the extracted features from N different drive trains of N conditions, where said trained neural network data is stored in unit (3.5) of the computing device (CD, CD′) and next generating an output vector (Z=[Z1, . . . ZN]), wherein the vector (Z) has N elements and if the output vector (Z) has only one element having value equal to 1 and the rest of elements values are equal to 0, determining the index (i) of the element with the value equal to 1, which index (i) indicates that the condition of she drive train under analysis is equal to the ith Condition of the N available known drive train conditions.