The fault diagnosis technology of the valve is to predict the health status of the valve by measuring the status information of the valve during operation, processing and analyzing the measurement data, combining the historical status information of the valve, and determining the necessary maintenance strategy for valves. The main research content involved in this technology includes the acquisition and transmission of operating status signals of valves, processing of status signals of valves and extraction of fault characteristics, mechanism of valve failures, and formulation of maintenance strategies of valves.
 
Since the nuclear accident in the Three Mile Island in the United States, the American Electric Power Research Institute (EPRI) has carried out fault diagnosis for valves for the first time at Duke Electric Power's Marshall Steam Station. The International Atomic Energy Association (IAEA) also gave guiding opinions about the fault diagnosis of valves used for nuclear power plants in the nuclear power plant's condition based maintenance document issued in 2007. Since then, extensive research on fault diagnosis technology of valves has been carried out at home and abroad, and there have been many achievements in all aspects.
 
Signal acquisition and sensing
Signal acquisition and sensing are the prerequisites for fault diagnosis for valves. For control valves, the diagnostic signal is generally obtained through smart positioners. Through the valve positioner, parameters such as stroke control signals, friction force, response time, spring stiffness and I/P conversion of valves can be obtained. The health status of the valve can be judged by the changes in these parameters. For on-off valves without a positioner, you need to install various sensors, such as attaching a strain gauge to the valve stem to measure the torque and thrust of the valve stem. In some cases where it is inconvenient to attach the strain gauge, a sensor with pincerlike thrust with low accuracy is used to measure the thrust of the valve stem. To measure the electrical parameters of the valve electric head, it is often necessary to turn on the electric head before connecting sensors such as voltage, current, travel switches, torque switches and so on. A pressure collection port on the pipeline should be provided in advance for the measurement of air pressure of pneumatic valves. Because strain gauges are expensive and difficult to install, many scholars indirectly calculate the stem torque by measuring the electrical parameters of the electric head, which is called the MCC (Motor Control Center) method. Kang and others evaluated the accuracy of this indirect measurement method. Although there was a certain error compared with the direct measurement, it could accurately reflect the change trend of moment of force and thrust of the valve in the opening and closing process. Jung and others improved the MCC method, and the test results showed that the estimation error of thrust of the valve was about 8.8% by the improved method.
 
In addition to detecting whether the valve is closed well, acoustic emission detection was the most commonly used for testing internal leakage of the valve. After the pressure bearing valve leaks, the medium impacts the valve body to make an acoustic emission signal. This signal is a continuous signal, and its RMS value is proportional to the leakage rate under certain conditions. However, this measuring method also has certain limitations. For example, when the pressure is not high, the signal is relatively weak, and the noise is relatively great. The quantitative relationship between the RMS value and leakage rate is obtained under the condition of knowing the installation position of the sensor and the position of the internal leakage. However, the position of the valve's internal leaks in actual working conditions is random. Therefore, there are certain limitations for acoustic emission technology for the current quantitative detection of internal leak in practical applications. Xinxin Wang studied the best installation position for detecting internal leakage using acoustic emission sensors for different valves, specifications and working media.
 
For valves with high medium temperature, the internal and external leakage and insulation materials of the valve are also measured by an infrared thermal imager at the industrial site. However, the detection of leakage of valves is only a qualitative method. Shitao Guan, Yichang Chen and Qijuan Chen researched the networking and remote signal transmission technology for fault diagnosis for valves.
 
The processing and analysis of the signal
Sensor signals collected on site often cannot be used to directly analyze the state of the valve due to various noises. The characteristic signal must be extracted from the original signal, which is a necessary condition for fault diagnosis of the valve. Signal amplification and filtering is the most commonly used method. In many cases, the original signal obtained by the sensor is very weak, such as the output of the strain gauge bridge and acoustic emission sensor. The original signal must be amplified before subsequent acquisition and analysis. Because the noise in the original signal is also amplified, the amplified noise must be filtered out. The signal obtained by the sensor sometimes cannot reflect the fault characteristics in the time domain. Fast Fourier Transform (FFT) is often used to transform the time domain signal into the corresponding frequency domain signal. Meland and others used spectrum analysis to analyze the internal leakage's acoustic emission signal of the cut-off ball valve and found that the signal had a distinct characteristic frequency in the frequency domain when the internal leakage of the ball valve occurred. As an effective signal processing method, wavelet packet transform has been widely used in the analysis and processing of stable signals and unstable signals. Zhang Haifeng and others used the wavelet packet method to decompose the acoustic emission signal of internal leakage of natural gas pipeline ball valves. The binary transform was especially used to re-decompose the high-frequency section after each layer was decomposed, which effectively compensated for the limitation of poor partial decomposition of the high-frequency section in the wavelet transform. Yuming Zhao and others used wavelet packets to decompose and reconstruct the vibration signal of the reciprocating pump valve to construct an energy feature vector, which could effectively reflect the failure characteristics of the reciprocating pump valve.