Large bearing heater
When testing under harsh environmental conditions such as high temperature, high noise, dust, vibration, etc., it will not only cause great harm to the physical and psychological of the inspector, but also make the inspector often unable to work normally. Therefore, the research on the detection of surface defects of bearing rings of large bearing heaters has become a hot spot in recent years. Based on digital image processing technology, our department has conducted research on the detection of surface defects of bearing rings of large bearing heaters. The main contents are as follows:
1. Typical performance type and defect area analysis of surface defects of bearing rings of large bearing heaters.
2. Analysis of image edge detection algorithm. A variety of classic edge detection operators are used to compare and detect the surface defect images of bearing rings of large bearing heaters, and an improved Sobel edge detection operator is proposed.
3. Extraction and selection of defect features. Hu defect invariant features, morphological features, and texture features were extracted from the defect image, and systematic analysis and demonstration were carried out to determine the Hu moment invariant features required for classification recognition.
4. Research on classification and recognition algorithm based on BP neural network.
Study on audio diagnosis method of bearing heater bearing fault
(1) The audio signal of the bearing heater bearing contains important information on its running status. By analyzing this information, the fault diagnosis of the bearing heater bearing can be performed effectively, and the audio signal can be collected in a non-contact manner, which is convenient to use and low in cost advantage.
(2) According to the advantage that all parameters in the Discrete Hidden Markov Model (DHMM) are discrete values, we propose a new method for audio diagnosis of bearing faults based on DHMM, which has simple modeling, fast calculation speed and diagnostic accuracy Advanced features.
(3) Since the continuous Gaussian mixture density function can be used to describe the output probability more reasonably, the paper proposes a new method of bearing fault audio diagnosis based on the continuous Gaussian mixture density HMM (Contlnuous Gaussian Mixture Hidden Markov Model, CGHMM). At the same time, the training and diagnosis algorithm is improved by using the cluster parameter-based model parameter initialization method and the calibration coefficient forward-backward algorithm.
(4) conducted a comparative analysis of the diagnostic test results of DHMM and CGHMM methods. The DHMM algorithm is better than the general CGHMM algorithm in speed, but the diagnostic accuracy is lower than the CGHMM algorithm.