Fitter technician's thesis

Misdiagnosis and information processing methods in mechanical fault diagnosis

Misdiagnosis of mechanical state is a distorted reflection of mechanical state. There are many reasons for misdiagnosis, including inaccurate diagnostic data, unreliable diagnostic basis and unreasonable diagnostic reasoning. The information characteristics of mechanical state play an important role in mechanical fault diagnosis, and it is of great practical significance to study these information characteristics to improve the accuracy and reliability of fault diagnosis. Aiming at the uncertainty of fault information obtained, a mathematical method theory of using rough set theory to deal with the uncertainty in diagnosis is proposed.

Keywords: fault diagnosis; Misdiagnosis; Information is unreliable; study

In the development of mechanical fault diagnosis, improving the fault diagnosis rate has always been the focus of research, but the misdiagnosis of faults has not attracted enough attention. In order to systematically explain the misdiagnosis in mechanical fault diagnosis, the meaning and classification of misdiagnosis are given. According to the process of mechanical fault diagnosis and reasoning, the mechanism and specific reasons of misdiagnosis are analyzed in detail, and the methods and measures to reduce misdiagnosis are put forward in view of these potential reasons. In order to improve the reliability of mechanical fault diagnosis and reduce the misdiagnosis rate, the diagnosis system must be reasonable, open and extensible while ensuring the accuracy of diagnosis data, so that the diagnosis knowledge can be enriched and enriched continuously.

1 analysis of the causes of mechanical misdiagnosis

Judging from the objective differences between the diagnosis results and the diagnosis objects, the conclusions of fault diagnosis can be divided into diagnosis, misdiagnosis and missed diagnosis, and diagnosis means that the fault judgment of the diagnosis objects is accurate. Missed diagnosis is the omission of fault. Misdiagnosis, as the name implies, is a wrong diagnosis, which can also be called misjudgment. Missed diagnosis can also be classified as equipment misdiagnosis.

1. 1 Complexity of fault

In the process of fault diagnosis, the fault process of the diagnosis object is complex and changeable. In the process of fault development, due to the different nature, characteristics and modes of action of the factors causing the fault, the specific conditions of mechanical action and damage are also different, and the symptoms and evolution of the fault have different forms, so it is often difficult to quickly and accurately understand the nature of the fault in diagnosis, leading to misdiagnosis, which is embodied in the following aspects:

During the development of (1) fault, a fault may show many different fault symptoms. For example, in the fault diagnosis of hydraulic system, the fault of electromagnetic directional valve may lead to the system pressure and flow rate not meeting the requirements, the pulsation may be intensified, and the working temperature of the system may also rise. However, for different diagnosis objects, even the same machine, the response to the same fault is different. One object may react quickly, another object may react slowly, one object's symptoms may react violently to faults, and the other object may react smoothly.

(2) In the development process of different faults, similar symptoms may appear, and the same symptom may correspond to multiple fault forms. For example, in rotating machinery, various faults are often accompanied by the intensification of vibration, while in frequency domain analysis, different faults may have similar performances at the same frequency doubling. The similarity of these fault symptoms makes us easily confused in fault diagnosis.

(3) In many cases, with the development of faults, it may also cause secondary faults, which may cover up the original faults or cover up the secondary faults, which will bring difficulties to fault diagnosis. For example, in the hydraulic system, the degree of oil pollution increases for some reason, which may cause serious wear of the moving pair of the hydraulic pump, and the wear particles are mixed into the oil, further aggravating the oil pollution. The wear of hydraulic pump will cause the failure of hydraulic system, which is caused by the primary failure of oil pollution. The primary failure and secondary failure of pump wear are mixed and promote each other, forming a vicious circle, which increases the difficulty of finding the primary failure.

In order to overcome the difficulty of fault diagnosis caused by the complexity of fault symptoms, we must broaden our thinking, not stick to the narrow thinking of typical fault symptoms, and make a concrete analysis from the system point of view, from environment to machinery, from part to whole, from stage to process, and organically combine symptoms, causes and fault mechanisms to reduce the misdiagnosis rate.

1.2 Uncertainty of diagnostic knowledge

Due to the different complexity and working environment of various mechanical equipment, our understanding of faults is often uncertain and imperfect. Generally speaking, we can't wait until a fault completely occurs before drawing a conclusion, but we must implement early diagnosis and take timely measures to avoid further development of the fault. In this way, we must make a diagnosis according to some symptoms or no symptoms of the fault, which will inevitably lead to misdiagnosis.

Due to the lack of fault diagnosis data, the understanding of the fault is greatly limited, and it is difficult to diagnose it clearly. Sometimes faults with similar symptoms cannot be completely ruled out, and sometimes the general law of suspected faults is not completely consistent with the symptoms of faults. In addition, the possibility of fault is ruled out, and there is also a lack of sufficient basis for fault identification. Therefore, the reasoning process of fault diagnosis often has certain fuzziness and uncertainty.

In view of this situation, it is more in line with the essence of fault diagnosis and will improve the reliability of diagnosis by fully studying the fault diagnosis object, establishing a reasonable fuzzy knowledge body and fuzzy reasoning machine, and applying modern artificial intelligence principles to carry out diagnosis.

The relativity of 1. 3 theory

Compared with the actual fault process, any theory always has limitations. Mechanical equipment, as an organism composed of environment and people, is different. Theory can only roughly summarize the specific situation in fault diagnosis practice. At the same time, the theory is limited by certain scientific and technological conditions, and there is still room for understanding.

There is always a certain distance between theory and specific faults. As far as the standard of fault diagnosis is concerned, it is summarized and formulated on the basis of typical symptoms. Untypical faults may not all meet the diagnostic criteria. If the diagnostic criteria are regarded as dogma and unchangeable, it will inevitably lead to misdiagnosis.

In a word, the diagnostic system we developed should be open and extensible, so that the system has the ability of continuous improvement, which is an important way to reduce the misdiagnosis rate.

1.4 limitations of diagnostic practice

Fault diagnosis practice is the foundation of the formation and development of mechanical fault diagnosis. Although it is also an important way to acquire relevant knowledge, due to the difference between the experiment and the actual operation and working environment of the mechanical system, the conclusions drawn must have certain limitations. As the main body of fault diagnosis, people have different understanding of mechanical system and practical experience of fault diagnosis, and come to different conclusions. For example, observing 1 machine image, 1 experienced person has accumulated a lot of fault knowledge in his mind, so he can often accurately grasp the running state of the machine and make reasonable diagnosis conclusions, especially for early faults and atypical faults. Therefore, we must strengthen the diagnosis practice and extract useful knowledge from the practice, so as to expand and enrich our diagnosis system.

The data obtained by 1.5 is inaccurate.

In the process of fault diagnosis, the relevant data of mechanical system operation should be obtained first. In the process of mechanical operation, it is often influenced by external environment and various random factors, which makes the obtained data inaccurate to some extent and easy to cause misdiagnosis. Therefore, it is necessary to take necessary data preprocessing measures to reduce the influence of random factors, eliminate trend items and singular items, and improve the accuracy of data, which is also a necessary condition to reduce the misdiagnosis rate.

1.6 The diagnostician is unprofessional.

The quality of the diagnostician also determines the correctness of the diagnosis conclusion. Diagnostic personnel's theoretical knowledge, practical experience, method knowledge and attitude in fault diagnosis may lead to misdiagnosis. At the same time, the diagnostician's ability to comprehensively apply knowledge, integrate theory with practice and be good at solving practical problems will also affect the diagnosis conclusion.

2 information extraction in mechanical fault diagnosis

2. 1 Information extraction is unreliable.

Mechanical fault diagnosis can be divided into direct diagnosis and indirect diagnosis, but due to the limitation of equipment structure and working conditions, direct diagnosis is often difficult to carry out. Therefore, indirect diagnosis is often used, that is, the state changes of key components in equipment are indirectly judged through secondary diagnosis information. Diagnostic test is the key link to obtain secondary diagnostic information. The most common are vibration test (displacement, speed, acceleration) and sound test.

However, due to various reasons, the data obtained may be biased. It is reflected in three aspects: (1) the data does not correctly reflect the objective existence; (2) The signal-to-noise ratio of the data is low; (3) The data is incomplete. If these inaccurate data are analyzed into valid data, it is likely to be misdiagnosed.

2.2 Information processing is not accurate

It is the key to extract the features that reflect the machine fault information quickly and effectively. Diagnostic features are mainly obtained by analyzing and processing the signals collected by the equipment. These features may be some simple time domain features, such as peak-to-peak value, root mean square value, kurtosis and so on. Or process parameters, such as oil temperature and oil pressure, as well as some complex frequency domain features and holographic spectrum-based features, such as variable frequency ellipse and axis trajectory.

At present, various feature extraction methods emerge one after another, such as statistical simulation, wavelet analysis, independent component analysis, frequency domain analysis and holographic spectrum analysis, which provide effective solutions for feature extraction of diagnostic objects. In application, many methods have their preconditions for application. In addition, in different applications, various methods may have their limitations and mathematical accuracy problems. In practical application, if we don't pay attention to these, it may cause misdiagnosis.

2.3 information is not perfect

For a diagnosis object, if its running state is complex, due to the limitations of objective conditions and means, it may be difficult to accurately give a diagnosis conclusion from the obtained information, mainly in the following three aspects:

(1) Information is incomplete. In the practice of diagnosis, there is not a one-to-one correspondence between faults and diagnostic information. 1 message corresponds to multiple different faults, and 1 fault is also characterized as multiple different messages. This requires enough useful information to distinguish different faults. Otherwise, misdiagnosis may occur.

(2) Inconsistent information. Inconsistency of diagnosis information is also a common phenomenon in diagnosis practice. There is a certain degree of conflict between these messages. That is to say, some information supports fault F 1 to a great extent and denies fault F2; On the contrary, other information supports fault F2 and denies fault F 1. Misdiagnosis is also prone to occur at this time.

(3) The information is uncertain. The diagnostic information from the diagnostic object passes through multiple transmission paths, and its uncertainty may be small or large, such as sensors and transmission lines, which affect its certainty. In addition, there is uncertainty caused by the transformation between qualitative and quantitative information.

3 Measures to improve information reliability and reduce misdiagnosis

3. 1 Improve the accuracy of diagnostic tests

Improving the accuracy of diagnostic test is an important prerequisite to ensure the reliability of diagnostic data. We can start from the following four aspects: (1) Check the sensor regularly; (2) Multiple sensors can be considered for measurement; (3) Adopt reliable transmission lines; (4) Set the sampling parameters correctly.

3.2 Improve the reliability of the diagnosis system

With the need of equipment operation and maintenance, various online, offline, remote diagnosis and analysis systems and intelligent diagnosis systems such as artificial neural network, Bayesian network and expert system are gradually used in mechanical fault diagnosis, which brings a lot of convenience to fault diagnosis and increases the possibility of misdiagnosis of mechanical faults. Developing a reasonable, perfect and effective diagnosis system to improve its reliability in feature extraction or diagnostic reasoning is conducive to reducing the misdiagnosis rate.

3.3. Strengthen the objectivity of diagnosis information description.

The importance of diagnostic information in mechanical fault diagnosis is self-evident, and whether its expression and description are reasonable and accurate is related to the correctness of diagnostic reasoning results. However, in the practice of diagnosis, there are both qualitative and quantitative information. It contains both simple information and complex information; There are both definite information and uncertain information. In the process of diagnosis and reasoning, quantitative information is often transformed into qualitative information, for example, the vibration of "70} and M is described as" big vibration "and so on.

Probability theory and fuzzy mathematics are powerful tools to describe this information. Therefore, probability theory and fuzzy mathematics theory can be considered to be integrated into the information expression and description of fault diagnosis in an appropriate way to strengthen the objectivity of its description.

4 Rough set theory to deal with information uncertainty

Information with one of random information, uncertain information, fuzzy information and grey information is simplex information; At least two kinds of information are blind, and the theories and methods of probability theory, fuzzy theory, grey mathematics and unascertained mathematics are organically combined, that is, the theories and methods of uncertain mathematics. Some theories and methods have been put forward or used to process information with similar or different feature patterns to obtain fused information, thus improving the uncertainty, fuzziness and contradiction of information.

Rough set theory is an effective mathematical tool to deal with fuzzy and uncertain knowledge, which has been successfully applied to knowledge classification and knowledge acquisition. The difference between rough set theory method and "soft science" methods such as neural network method, genetic algorithm, fuzzy set theory method and chaos theory is that it only uses the information provided by the data itself, and does not need any additional information or prior knowledge, such as basic probability assignment in evidence theory, membership function in fuzzy set theory and probability distribution in statistics. Rough set method directly processes the measurable output of the object on the basis of classifying the observation data, thus eliminating redundant information and contradictory information. General steps of diagnosis based on rough set theory;

(1) Establishment of knowledge base The fault information table of the joint diagnosis system is generated by using the collected historical or simulation data, and then expressed in the form of knowledge base S=(U, a, d, v, f) The knowledge base can be divided into non-empty object space U = 3x >；; & gtXz, …, xm (,attribute set space R=AV D, subset A=}dl} a2 }…, do} and subset D=}df are called conditional attributes and result attributes respectively, and there are object attribute relationships f8 and 9 between the object space and the latter two, that is, for. EA, F8:' inkstone lifting' is called information function, while' ‘er:U} Vd is called decision function, and Va and yd are limited value ranges of A and D respectively.

(2) Data discretization The data discretization methods include equidistant segmentation algorithm, equal frequency segmentation algorithm, NaiveScaler algorithm, attribute-based importance algorithm and breakpoint-based importance algorithm, and the algorithm combining Boolean logic with rough set theory. The values of conditional attributes and decision attributes are continuous uncertain spaces, and data discretization is data preprocessing using rough set theory.

(3) Feature extraction finds m data features from the original n data features, and the classification ability of the simplified m data features to the object space U is the same as that of the original n data features (n, m). This process is called feature extraction. Commonly used feature extraction methods include minimum reduction based on attribute importance, logical reduction based on discernibility matrix and discernibility function, maximum distribution reduction based on inclusion theory, attribute reduction based on lower approximate quality invariance and arbitrary reduction based on upper approximate quality in the presence of noise pollution. Feature extraction reduces conditional attributes, and then eliminates redundant conditional attributes.

(4) The rule set extracted by rule application can be used to classify new objects. This rule set is called "classifier" and expressed by RUL. When the classifier encounters a new object X, it looks for rules that match the conditional attributes of X in the rule set RUL, and can judge the decision attributes of the new object X by applying the rule set.

refer to

1 Qu Liangsheng, He. Mechanical fault diagnosis. Shanghai: Shanghai Science and Technology Press, 1986.

Liu Zhenhua, Chen Xiaohong. Misdiagnosis. Jinan: Shandong Science and Technology Press, 200 1.

3 Qu Liangsheng, Wu Songtao. Some applications of statistical simulation in engineering diagnosis. Vibration, test and diagnosis, 200 1, 21(3):157-167.

References:

Lifting and transportation machinery 2008