The boiler room is an important place for production, and it also causes frequent failures due to its frequent use. Common failures are caused by boiler scale, corrosion, removal, and bearing oil leakage and burning. Boiler failures are often extremely serious. Therefore, it is necessary to explore the prediction method of common boiler failures, and it is also an important part of ensuring the production and life safety of the people.

1. Knowledge about Boiler Failure Prediction

Artificial intelligence fault diagnosis and prediction technology is a new technology that has emerged with the rapid development of modern technology and economy. It can identify whether the state of the equipment is normal, find and determine the location and nature of the fault, and propose corresponding countermeasures to improve the quality of the equipment. Reliability of operation, prolonging its service life, and reducing the cost of the whole life cycle of equipment. The use of fault prediction technology can realize the early detection of faults and predict their future development trends, which is convenient for timely adjustment of thermal power units to avoid malignant accidents, so that the units can operate safely and reliably, and at the same time improve the economy of the units.

According to the length of the forecast period, the failure forecast can be divided into three types.

Long-term forecasting is for making long-term maintenance plans and maintenance decisions for boiler units. The time is generally more than one month, and the forecasting accuracy is low.

Mid-term forecasting is to predict the state of the boiler unit in a relatively long period of time in the future, and serve the unit’s mid-term maintenance plan and maintenance decision-making service. The time is generally about one week, and the prediction accuracy is relatively low.

Short-term forecasting is to predict the short-term development of the boiler unit. The time is about one day, and the prediction accuracy is high.

For medium and long-term forecasting, due to the low accuracy requirements, a simple forecasting model can be considered to establish a univariate time series model for forecasting. For short-term forecasting, the accuracy requirements are higher, and because various relevant factors have a greater impact on the state value at that time, when making short-term forecasting, in addition to considering the time series itself, other relevant factors should also be properly taken into consideration. This requires the establishment of a multivariate time series model for forecasting to meet the accuracy requirements of short-term forecasting.

 2. Fault Prediction Accuracy Requirements

Today's intelligent fault diagnosis and prediction technology is a new technology emerging with the rapid development of modern technology and economy, because it can identify whether the state of the equipment is normal, discover and determine the location and nature of the fault, and propose corresponding countermeasures , in order to improve the reliability of equipment operation, prolong its service life, and reduce the cost of equipment life cycle. 

The use of fault prediction technology can also realize early detection of faults and predict their future development trends, which facilitates timely adjustment of the unit and related equipment, avoids the occurrence and expansion of malignant accidents, ensures safe and reliable operation, and improves the operation of the unit. economy. In terms of the forecast period, according to the length of the forecast period, fault prediction can generally be divided into: long-term forecast, medium-term forecast and short-term forecast.

2.1 Long-term forecast

It is a forecast for making long-term maintenance plans and maintenance decisions for boiler units, and the time is generally more than one month, and the accuracy of the forecast is low.

2.2 Medium-term forecast

It predicts the state of the boiler unit in a relatively long period of time in the future, and serves for the mid-term maintenance plan and maintenance decision of the unit. The time is generally about one week, and its prediction accuracy is also relatively low.

2.3 Short-term forecast

It predicts the recent development of the boiler unit, usually about one day in time, and its prediction accuracy is relatively high. For medium and long-term forecasts that do not require high precision, a simple forecast model can be considered to establish a univariate time series model for forecasting. For short-term predictions with high precision requirements, various relevant factors have a greater impact on the state value at that time. In addition to the time series itself, other relevant factors should also be properly taken into account when making short-term forecasts. Therefore, it is necessary to establish a multivariate time series model for forecasting to meet the accuracy requirements of short-term forecasting.

3. Commonly Used Boiler Failure Prediction Methods

In recent years, many researchers have used linear regression analysis, time series analysis, gray model prediction method, expert system, artificial neural network and other methods to carry out boiler equipment fault diagnosis research, in order to explore fast and effective fault diagnosis and prediction methods. The commonly used forecasting methods are:

3.1 Linear regression analysis method

Regression analysis is a method of statistical inference to find the mathematical relationship among several uncertain variables. The simplest of these relationships is linear regression analysis.

3.2 Time series analysis method

Time series refers to a set of data arranged in chronological order, and time series analysis refers to a data processing method that uses a parametric model to analyze and process the observed orderly random data. Time series analysis methods mainly include curve fitting, exponential smoothing, seasonal model, linear stochastic model, etc., which are mainly suitable for single-factor forecasting, while boiler failure forecasting has both deterministic trends and certain randomness. When predicting multi-factors, it is necessary to separate deterministic trends, and the calculation is more complicated. At the same time, it is necessary to assume the zero mean and stationarity of the separation residuals, and the prediction accuracy is not high.

3.3 Gray model prediction method

The gray model prediction method is to establish a prediction model according to the gray system theory. It establishes a general gray differential equation according to the general development law of the system, and then obtains the coefficients of the differential equation by fitting the data sequence, thereby obtaining gray prediction. model equation. There are two main methods to apply the gray system theory to fault prediction, one is the gray prediction model based on the gray system dynamic equation GM (or DM), and the other is the residual identification prediction model based on the residual information data column. Among them, the GM prediction model is a differential equation with 1 order and 1 variable, and the gray model described is more commonly used. From a mathematical point of view, the solution of gray forecasting is equivalent to the superposition of power series, which includes the content of general linear regression and power series regression, so the gray forecasting model is better than general linear regression or exponential curve fitting. in deterministic time series analysis.

3.4 Expert system

Expert systems can successfully solve problems in some specialized fields and have many advantages. However, after years of practice, it is always far from the level of experts, and sometimes it is not as good as a beginner on some issues. Analyzing the reasons, there are mainly the following aspects: the "bottleneck" problem of knowledge acquisition; the limitation of a single reasoning mechanism that simulates the expert's thinking process; the system lacks self-learning ability.

3.5 Artificial Neural Network Forecasting Method

There are many problems in the fault diagnosis of neural network. It cannot make good use of the experience and knowledge accumulated by experts in the field. It only uses some clear fault diagnosis examples, and it needs a certain number of samples to learn. After training, some threshold matrices and The weight matrix is not a logical reasoning production like expert experience knowledge, so it lacks the ability to interpret the diagnosis results, cannot be applied to real-time diagnosis, and can only process historical record data.

3.6 Combination of expert system and artificial neural network

The combination of expert system and artificial neural network is a hot research topic at present. The neural network expert system composed of neural network and expert system can use the characteristics of large-scale parallel distributed processing and knowledge acquisition automation of neural network to solve the "bottleneck" of knowledge acquisition, weak reasoning ability, poor fault tolerance, It is difficult to deal with large-scale problems, realize parallel association and adaptive reasoning, improve the intelligence level of the system, and enable the system to have real-time processing capabilities and high stability. 

Compared with the traditional expert system, the expert system based on neural network has the following advantages: it has a unified internal knowledge representation, and any knowledge rule can be stored in each connection weight of the same neural network through the learning of examples. It is convenient for the organization and management of the knowledge base, and has strong versatility; the knowledge capacity is large, and a large amount of knowledge can be stored in a corresponding.

For a much smaller neural network; it is convenient for automatic acquisition of knowledge and can adapt to changes in the environment; the reasoning process is a parallel numerical calculation process, which avoids problems such as slow reasoning speed and low efficiency; it has image thinking such as association, memory, and analogy The ability to work beyond the scope of the learned knowledge; realize the integration of knowledge representation, storage and reasoning, that is, all are realized by a neural network.

Conclusion

With the development and progress of society, we pay more and more attention to the prediction method of boiler failure. The mastery of the prediction method of boiler failure is of great significance to real life. It is very important to maintain the normal operation of the boiler and ensure normal production. For the work of boiler failure prediction, it is necessary to carry out.

Further Reading:

Research on Energy Saving and Emission Reduction Strategy of Industrial Steam Boiler

How to Make Industrial Boilers Move Towards a New Stage of Automation and Intelligence?

Boiler Guidance| Industrial Boiler Operation Optimization and Adjustment Method