Condition monitoring technology for rotating machinery has been a hot research topic in recent years, with emphasis on avoiding random equipment failures. Compared with periodic monitoring and online detection with offline analysis, automatic online monitoring is technologically advanced, avoiding sudden equipment failures and requiring no on-site professional operation. Online prediction technology for rotating machinery conditions is one of the emerging research topics, with emphasis on predicting time-dependent equipment failures and changing equipment maintenance methods. This technology is developed on the basis of condition monitoring and fault analysis, and is the key technology to replace time-based preventive maintenance with advanced predictive maintenance. This project focuses on the methods of online condition monitoring and trend prediction for equipment.

　　1　Development of Condition Monitoring Technology for Rotating Machinery

　　1.1　Development History of Condition Monitoring Technology for Rotating Machinery

　　Rotating machinery is the most widely used machinery in industry. Many large rotating machines, such as centrifugal pumps, motors, engines, generators, compressors, steam turbines, rolling mills, etc., are key equipment in petrochemical, power, metallurgy, coal, nuclear energy and other industries. Since the beginning of this century, with the rapid development of the machinery industry, mechanical equipment in modern mechanical engineering has developed towards light weight, large scale, heavy load and high automation. A large number of problems concerning strength, structure, vibration, noise, reliability, as well as materials and processes have emerged, and equipment damage incidents occur from time to time. Serious accidents of large steam turbines at home and abroad are typical examples.

　　The research on condition monitoring technology for large rotating machinery is a national key project, aimed at improving the product quality of large rotating machinery, reducing sudden accidents, and avoiding major economic losses. In the 1950s, various types and performances of sensors and vibrometers were successively developed and began to be applied in scientific research and engineering practice. In the 1960s and 1970s, the development of digital circuits and electronic computer technology and the formation of "signal digital analysis and processing technology" promoted the application of vibration detection technology in mechanical equipment. From the 1970s to the 1980s, research on condition monitoring and fault diagnosis technology for mechanical equipment began in many developed countries. With the development of modern science and technology such as electronic computer technology, modern testing technology, signal processing technology, signal recognition technology and fault diagnosis technology, the monitoring research of mechanical equipment entered a systematic stage, and the laboratory research results were gradually extended to nuclear energy equipment, power equipment and other large complete sets of mechanical equipment, entering a stage of vigorous development. For example, the "Machinery Health Monitoring" (MHM) system of Mitsubishi Corporation of Japan, the "Mobile Diagnosis Center" (MDC) of Westinghouse Corporation of the United States, and the 2500 vibration monitoring system of B&K Corporation of Denmark all have functions such as acquisition, analysis, calculation, display, printing and plotting of unit signal data, and are equipped with special diagnostic software. Advanced condition monitoring systems take vibration and noise, which reflect the dynamic characteristics of machinery, as the main content of monitoring and analysis. Since vibration and noise are fast random signals, not only high requirements are placed on the test system, but also a large amount of data processing is required in the analysis. In the 1980s, at home and abroad, small computers or special digital signal processors were used as host machines to complete data processing of mechanical dynamic characteristics (such as HP5451C). Such host machines are not only expensive (generally hundreds of thousands of yuan) but also have harsh requirements on the working environment (requiring a dedicated computer room), so offline monitoring and analysis methods are usually adopted.

　　Since the 1990s, high-end microcomputers have been continuously updated and their prices have dropped rapidly, and calculation methods suitable for digital signal processing have been continuously optimized, greatly improving data processing speed and creating conditions for the direct application of condition monitoring technology in industrial sites. Experts and scholars from developed countries such as Denmark, the United States, Germany, and Japan have conducted in-depth research on condition monitoring technology for rotating machinery and developed different systems. Such systems, represented by the 2520 vibration monitoring system of B&K Corporation of Denmark, the 3300 series vibration monitoring system of BENTLY Corporation of the United States, and the M6000 system of Atlanta Corporation of the United States, have reached a relatively high level. One of the functionally typical systems is the 2520 vibration monitoring system (vibration monitor-type 2520) of B&K Corporation of Denmark, whose main functions include: automatic spectrum comparison and fault warning and alarm; speed compensation for 6% and 23% constant percentage bandwidth spectra; amplitude growth trend graph display; three-dimensional spectrum display; calculation of total vibration root mean square value (vibration severity); support for local area networks. The IQ2000 system of IRD Corporation of the United States can be regarded as the most fully functional monitoring and diagnosis system reported so far.

　　China's research on condition monitoring technology in industrial sectors started in 1986. Before that, large units introduced from abroad were generally equipped with monitoring systems. However, for self-developed domestic equipment, if foreign monitoring systems are selected, they are difficult to accept due to extremely high prices. Since the mid-to-late 1980s, relevant research institutes, universities and enterprises in China have begun to independently or cooperatively research condition monitoring technology for rotating machinery, and have achieved results in theoretical research, testing technology and instrument development, and developed corresponding condition monitoring systems for rotating machinery, such as those by Xi'an Jiaotong University, Zhejiang University, Beijing Institute of Technology, Beijing Institute of Machinery Industry, etc.

　　There are mainly several types in China: a. 3MD-Ⅰ, 3MD-Ⅱ, 3MD-Ⅲ systems jointly developed by Harbin Institute of Technology and other units; b. RMMDS system of the Mechanical Monitoring and Diagnosis Research Office of Xi'an Jiaotong University; c. RB20-1 system of the Lubrication Theory and Bearing Research Office of Xi'an Jiaotong University; d. RMMDS system of Zhengzhou Institute of Technology; e. CDMS system of Chongqing Taidi Company; f. CMD-I and II systems of Zhejiang University; g. MD3905 system of Northwestern Polytechnical University; h. BJD-ZⅠ, BJD-ZⅡ, BJD-ZⅢ systems of Beijing Institute of Machinery Industry. Among them, the more typical systems include: the 3MD-Ⅰ microcomputer-based "Steam Turbine Generator Set Vibration Monitoring and Fault Diagnosis System" jointly developed by Harbin Institute of Technology and other units and passed appraisal in October 1985, and the subsequently further developed steam turbine fault diagnosis expert systems 3MD-Ⅱ and 3MD-Ⅲ; the RMMD-S "Microcomputer Condition Monitoring and Fault Diagnosis System" for five major fertilizer units developed by the Mechanical Fault Diagnosis Research Office of Xi'an Jiaotong University and passed appraisal in 1987. The main functions of these systems include: shaft vibration monitoring, including shaft center trajectory analysis, axial displacement, and calculation of shaft vibration displacement peak-to-peak value; casing vibration monitoring; spectrum analysis, including frequency zoom, order spectrum analysis, order tracking spectrum, three-dimensional power spectrum analysis; automatic pre-alarm and alarm; fault feature extraction and diagnosis.

　　The software functions of the above systems are relatively rich, and the hardware performance is also continuously improved, but they are basically still in the research and development stage, and the prices remain expensive. These systems are mainly used for monitoring key equipment in national key enterprises or specific equipment, such as large steam turbine units and large hydro-turbine units.

　　From the perspective of technological development, modern monitoring technology has roughly gone through two stages.

　　The first stage is the conventional technology development stage based on sensor technology and dynamic testing technology, with signal processing technology as the means. The technology of this stage has been applied in engineering. It has absorbed a large number of modern scientific and technological achievements. The leap-forward development of sensor technology has made it possible to use various information such as vibration, noise, force, temperature, electricity, magnetism, light, and rays. Thus, monitoring and fault analysis technologies such as vibration, noise, spectrum, ferrography, non-destructive testing, and thermal imaging of equipment have emerged. The development of signal analysis and numerical processing technology, combined with the development of microcomputer technology, has brought about various methods, such as state space analysis, comparative analysis, function analysis, logic analysis, statistical and fuzzy analysis methods. In recent years, the emergence of various data processing software and hardware has made real-time online monitoring and fault analysis technology possible.

　　Artificial intelligence technology provides the possibility for the intelligent development of equipment monitoring and fault analysis, bringing modern monitoring technology to the second stage. The research content and implementation methods of this stage have begun and are continuing to undergo major changes. The process centered on data processing will be replaced by a process centered on knowledge processing. Research on theories, methods and application technologies of expert systems, neural networks and fuzzy analysis has been carried out. What will play a leading role in this stage will be the knowledge of human experts, including the domain knowledge possessed by human experts and the methods for solving problems. Due to the unification of signal detection, data processing and knowledge processing, advanced technology is no longer a technology that only a few professionals can master, but a tool that general operators can use.

　　1.2　Development Trends of Condition Monitoring Technology for Rotating Machinery

　　The monitoring technology for the operating status of mechanical equipment has evolved from mere intuitive listening, looking, and touching to advanced monitoring technology using modern measurement technology, computer technology and signal analysis technology, such as ultrasound, acoustic emission, infrared temperature measurement, etc., which emerge in an endless stream. Emerging disciplines such as artificial intelligence, expert systems, and fuzzy mathematics have also found their place in mechanical condition monitoring technology.

　　In the analysis methods and application technology of mechanical dynamic signals, recent developments include: preprocessing using spatial domain filtering, multi-axis order signal analysis technology using Vold-Kalman filtering, Wigner-Ville distribution analysis suitable for non-stationary signals, wavelet transform methods, chaos analysis methods, intelligent sensing and detection technology, and technologies related to the VXI bus instrument platform, etc.

　　Nowadays, there are mainly three typical condition monitoring modes at home and abroad.

　　(1) Offline periodic monitoring mode. Test personnel regularly go to the site and use a sensor to test each measuring point in turn, and record signals with a tape recorder. Data processing is completed on a dedicated computer or directly on a portable instrument with a built-in microcomputer; this is the mode commonly used with imported monitoring instruments at present. With this mode, the test system is relatively simple, but the testing work is cumbersome and requires specialized test personnel; because it is offline periodic monitoring, sudden failures cannot be avoided in time.

　　(2) Online detection with offline analysis monitoring mode. Also known as the master-slave monitoring mode, sensors are installed at multiple measuring points on the equipment, and the on-site microprocessor slave system performs data acquisition and processing for each measuring point, while professionals analyze and judge on the host system. This mode has been adopted for large rotating machinery in recent years. Compared with the first mode, this mode eliminates the trouble of changing measuring points and can perform online detection and alarm; however, this mode requires offline data analysis and judgment, and the analysis and judgment require the participation of professional technicians.

　　(3) Automatic online monitoring mode. This mode can not only automatically monitor the working status of equipment online and provide timely fault prediction, but also perform online data processing and analysis and judgment; because it can perform intelligent comparison and judgment based on expert experience and relevant criteria, duty staff with a secondary education level can use it after short-term training. This mode is the most technologically advanced, requires no manual replacement of measuring points, and requires neither specialized test personnel nor professional technicians to participate in analysis and judgment; however, the workload for developing software and hardware is large. This project studies this mode.

　　In the future, condition monitoring technology for rotating machinery will tend to develop from offline periodic monitoring and online detection with offline analysis monitoring to automatic online monitoring. With the development of artificial intelligence theory and its practical application, and the massive development of data processing software, condition monitoring technology for rotating machinery will develop towards multi-objective, multi-level monitoring and networking in the future.

　　2　Development of Condition Prediction Technology for Rotating Machinery

　　2.1　Development History of Condition Prediction Technology for Rotating Machinery

　　When mechanical equipment fails, not only material wealth is destroyed and services are forced to be interrupted, but even the survival of personnel may be threatened. In industrial history, disasters and environmental accidents caused by mechanical equipment failures have occurred frequently. For example, the crude oil leakage accident of the Amoco Cadiz oil tanker in the United States, the Chernobyl nuclear power plant accident in the former Soviet Union, etc. Understanding the process of these accidents and how to prevent them has become an important issue to consider. In particular, most of these failures are caused by human intervention and improper measures, so reducing the number of maintenance operations and improving the scientific nature of maintenance are important aspects of preventing vicious accidents.

　　Traditional mechanical equipment maintenance methods can be summarized as: run-to-failure maintenance and time-based preventive maintenance; the former is generally used for cheap small machines, using backup equipment to ensure production; the latter, also called periodic maintenance, is generally used for large and medium-sized equipment, and is regularly overhauled according to artificially planned time regardless of whether the equipment has faults. Since the 1980s, a comprehensive engineering discipline has been formed with the goal of establishing a new maintenance system, and this engineering discipline has been promoted in different forms in Europe, America, and Japan. In recent years, experts and scholars from developed countries such as Denmark, the United States, Germany, and Japan have further proposed the basic concept of predictive maintenance. Since the 1990s, research has begun on new types of rotating machinery working condition analysis and condition prediction technology, and on the application of new technologies such as expert systems and neural networks. However, research on artificial intelligence online condition prediction and predictive maintenance is still in the initial stage of research and development.

　　Equipment predictive maintenance replaces periodic overhaul by monitoring and predicting the operating status of mechanical equipment. Its principle is: maintenance is carried out only when the monitoring, analysis and prediction results indicate that maintenance is necessary. This modern maintenance method can monitor and predict equipment failures, shut down the equipment in time when signs of failure are found, and even carry out targeted maintenance according to the identified nature and location of the fault. Its detection method is usually periodic detection, but the ideal method is online real-time detection; its analysis and prediction method is usually completed by professionals on a computer, but the ideal method is completed by an artificial intelligence system in real time online.

　　Therefore, if the past and current states of mechanical equipment can be detected online in real time and analyzed by artificial intelligence, and the subsequent development can be predicted, then the current working state of the mechanical equipment can be scientifically and effectively revealed at any time, and it can be predicted how long it will take for the equipment state to reach an unacceptable level and require shutdown for maintenance, thus upgrading from traditional preventive maintenance to predictive maintenance. To implement predictive maintenance for rotating machinery equipment, it is necessary to further realize it through comprehensive analysis and judgment of equipment status in frequency domain and time domain, as well as trend prediction of status, on the basis of condition monitoring and fault analysis of rotating machinery.

　　An internationally representative prediction system is the preventive maintenance system of Entek Scientific Corporation of the United States. Its main functions include: amplitude trend graph display; time-domain waveform display, spectrum display; automatic alarm for six frequency bands, automatic alarm for narrow-band spectrum; display of amplitude ratio of two spectra, display of amplitude difference of two spectra; three-dimensional spectrum display; offline fault diagnosis using a rotating machinery fault diagnosis expert system; support for ferrography analysis; support for local area networks. This prediction system can automatically compare spectra, identify frequency drift caused by changes in rotating machinery speed, and provide alarm signals.

　　With the development of science and technology in China, some large enterprises are gradually transitioning from simple vibration measurement or patrol inspection, periodic detection and overhaul to long-term continuous monitoring and predictive maintenance. The research directions of some universities and research institutes have also begun to change accordingly. A representative one is the research on the Windows-based IDPM intelligent diagnosis and predictive maintenance software system of Tianjin University. However, the current research focus in China is still concentrated on condition monitoring and fault analysis of rotating machinery equipment, while the intelligent online condition prediction technology aimed at predictive maintenance for large rotating machinery units still needs systematic research. Most of the monitoring systems developed by many domestic manufacturers and research units have single measurement items, and are even often limited to the detection of conventional parameters such as temperature, pressure, liquid level, and electrical quantities. They do not have the function of detecting and analyzing mechanical dynamic characteristics mainly based on vibration, and therefore cannot reflect the important working status of rotating machinery equipment; even if they have the function of detecting vibration, they are still limited to condition monitoring and fault analysis and cannot predict the development trend of the working status of rotating machinery equipment.