According to the content of data analysis, PHM can be divided into six topics: sensor data processing, condition monitoring, health management, fault diagnosis, fault prediction and operation and maintenance optimization. These topics are based on the monitoring data of equipment and environment and the full-dimensional historical data of equipment (called equipment archive model here). The PHM analysis framework is shown in Figure 4-4. The differences between this framework and OSA-CBM system [8] are as follows: ① Fault diagnosis and fault prediction are separated, with fault diagnosis after the event and fault prediction before, and their analysis algorithms are obviously different; ② In data collection, monitoring data and equipment management data are separated, which are usually independent in IT systems and have different data modeling techniques.
Condition monitoring can find abnormal state in time according to the latest condition monitoring information. Compared with the classical threshold alarm of SCADA and DCS, PHM has to deal with complex alarm rules such as abnormal pattern detection and abnormal trend identification based on multiple time series, eliminate false alarm according to multi-sensor fusion and trend analysis, merge alarm information according to fault propagation mechanism, and eliminate alarm shower. Sometimes the statistical analysis method can also provide more suitable threshold estimation for the threshold alarm rules of SCADA and DCS adaptively.
Fault prediction is to predict the remaining life, fault time and fault risk of equipment based on fault symptom indication and equipment degradation process modeling. When a system, subsystem or component may have minor defects and early failures or gradually degenerate to a state where it cannot complete its specified functions with the best performance, these minor defects, early failures or performance degradation can be detected by selecting relevant detection methods and designing a prediction system, so that equipment maintenance personnel can predict the failure time and take a series of preventive maintenance measures without waiting for the failure to make a passive response. Fault symptoms refer to anomalies that can be observed before a fault mode occurs or at the initial stage of its development.
Health management refers to monitoring the operation parameters of each system in various ways under the operation or working state of each system, and judging whether the system can work normally (task ability) under the current situation. Health assessment and diagnosis have opened up a new road for improving the reliability, maintainability and effectiveness of equipment. In order to avoid the paralysis of the whole equipment system caused by some operating processes, it is necessary to deal with the failures quickly, maintain normal basic functions, improve the utilization efficiency and safety of the equipment, and ensure safe and reliable operation.
The premise of fault diagnosis is to understand the fault mechanism of equipment. For example, abnormal vibration and noise signals of fluid rotating machinery such as steam turbine and compressor provide important information for fault diagnosis in time domain and frequency domain, which is only a part of fault information; The load of fluid machinery and the changes of medium temperature, pressure and flow rate have an important influence on the running state of equipment, which is often an important factor leading to abnormal and unstable operation of equipment. Therefore, for the fault diagnosis of rotating equipment, only on the basis of obtaining the steady-state data, transient data, process parameters and running state information of the machine, through the calculation of fault symptoms, the extraction of fault sensitive parameters and comprehensive analysis and judgment, can the cause of the fault be determined, the practical diagnosis conclusion be drawn, and the treatment measures be put forward.
On the basis of the above analysis, according to the characteristics and complexity of the system, appropriate methods can be adopted to realize operation and maintenance optimization. Combining the objectives of cost, time, efficiency and service life of equipment, and considering the constraints of resources, time window requirements and compliance requirements, logistics support activities such as maintenance personnel allocation and maintenance planning can be optimized. Generalized optimization also includes providing knowledge during operation and maintenance to improve maintenance efficiency.
In many scenarios, enterprises hope that data analysts can sort out the planning of intelligent operation and maintenance from the perspective of enterprise ecology and management, and form the problem of big data analysis of intelligent operation and maintenance. There are three main application modes of PHM, as shown in Table 4-9.
Recommended reading: "Industrial Big Data Analysis Practice" comprehensively understands the relevant knowledge of industrial data analysis.
Editor Yu Gang