Diagnosis of Mechanical Fault Based on Rough Set Theory

Abstract. As a kind of tools for dealing with fuzzy and not sure knowledge， rough set theory can point out effectively what knowledge is redundant and what knowledge is useful with the precondition of unchanged classification capacity. Therefore， the mechanical fault diagnosis based on traditional artificial intelligence has turned into the computational intelligence based on rough set theory. In this paper， taking an example of a type of mechanical fault diagnosis， it simplifies the fault diagnosis rules and achieves good effect to fault diagnosis with rough sets theory through the original data analysis and attribute reduction.

Keywords： rough set； attribute reduction； fault diagnosis

1. Introduction

With the quick development of Chinese agricultural modernization， agricultural machinery was put forward higher request and the intelligent system of large integrated agricultural machinery became more and more important. Featured complicated structure， higher accuracy and fast running speed are its typical characteristics. However， the more complex degree in the mechanical system， the more close relationship between the components， the more times the mechanical system breakdown. Generally speaking， it is called the mechanical system fault， that is， the mechanical system will partly or whole lost the working ability when parts or even entirety of a mechanical system work in abnormal state. Success in the fault diagnosis will have good control of the system， so as to ensure the control system can highly efficient and stable operation， [1] and to avoid the larger economic loss because of the amplification effect of the fault. Therefore， it is the key to improve agricultural production efficiency that diagnosis the fault in the first time and then keep the agricultural machinery equipment work with high efficiency.

At present， the technology of agricultural machinery fault diagnosis mainly use the expert system based on artificial intelligence. It contains the following core parts： the fault information knowledge base， knowledge acquisition， the reasoning machine and the explain part. [2] The most advantage of this theory is to imitate the human brain to fulfill the whole process of the fault diagnosis in the level of experts. It is a kind of the advanced analysis reasoning systems， not only to monitor the working state of agricultural machinery system， but to forecast the developing trend. However， it was filled with a lot of noise in fault information acquired in the actual work. It will form the incomplete fault information base and it is often not accurate to reason the conclusion according to this expert system with incomplete knowledge base. This is also the most serious defects of expert system for agricultural machinery fault diagnosis at present. Therefore， the problem the expert system has to solve is to overcome the negative factors， such as uncertain， inconsistent and incomplete， and to highlight fault feature， so as to realize accurate diagnosis.

Rough Set theory is put forward in 1982 by Z. Pawlak， a Polish mathematician. It is a kind of mathematical tool can analysis the imprecise， the inconsistent and incomplete information and knowledge. [3] This theory has been proved that it is suitable to simplified the data and correlation， search the similarity， find the data model and extract the reasoning rules without any priori information but the necessary data processing. [4-7] In fact， because processing data can work without any priori information， processed data information with the rough set theory can objectively reflect the problem described. Along with the improvement of the complexity in agricultural machinery system， mechanical system fault also presents the diversity and concealment. What’s more， the lag of the study in mechanism and the blind collection lead the large incomplete information without mechanical system fault diagnosis timely and effectively. The advantage of rough set theory is to solve this problem.

In order to enrich agricultural machinery fault diagnosis methods， this paper will study on the application of agricultural machinery fault diagnosis from the rough set theory of the attribute reduction.

2. THE MECHANICAL FAULT DIAGNOSIS METHOD BASED ON ROUGH SET THEORY

Nowadays， methods of the mechanical fault diagnosis are numerous， but incomplete and uncertainty of the description of the fault information because mechanical fault diagnosis process is nonlinear [8]. Thus， only the diagnosis method based on knowledge has the strong practical significance. The rough set theory can manage the incomplete and uncertain information under the guarantee of the classification ability unchanged. And it is inevitable for

rough set theory to apply to the mechanical fault diagnosis. So， fault diagnosis based on the rough set theory is the research focus， and its basic steps can be summarized as follows roughly [9] ：

Step 1： The invalid or deficient information because of the diversity of the original fault data acquisition way， so it is particularly important to pretreat the original data， including polishing missing information and deleting the ineffective information. In addition， attributes discrimination to the ones with continuous attributes fault decision table is needed， because processing object can only be the discrete data with using rough sets theory treatment decision table.

Step 2： To reduce the information system that is to remove the redundant attributes. Attribute reduction and the attribute value reduction are the most important contents， and the main method for rough sets theory to process the knowledge. The goal of attribute reduction is to ensure that mechanical fault classification ability unchanged at the same time and to remove unnecessary attribute， then keep the main reason attributes.

The reduction process in rough set theory is mainly processing with the premise condition of unchanged distinguish relationship between before and after. The final goal of this process is to get the smallest reduction attributes subset which has the same classification ability of initial attribute set. Delete algorithm of information system is the most use for reduction method. If new information forms do not distinguish between the same relationships with the original information table， the deletion is redundant. Delete algorithm are as follows：

（1） To delete any column attribute in decision table.

（2） To delete the same lines， namely duplicate examples.

（3） If the decision table can not distinguish the relationship unchanged， it should be deleted this attribute； If the decision table can not distinguish the relations changed， it should be kept.

Through the above attribute reduction process， the extraction time and work of fault diagnosis information can be reduced， so as to ensure the efficiency of the diagnosis.

Step 3： Not all the records of information table after the attribute reduction is necessary， so having attribute value reduction for the decision table after attribute reduction is needed. The specific practices are to analyze ear record， and then delete the redundant condition attribute values which are useless to mechanical fault classification. So each sample of decision table represents a certain kind of fault diagnosis decision rules.

3. APPLICATION EXAMPLE

The following is about the fault diagnosis analysis to a certain type of agricultural machinery equipment. The original data of the agricultural machinery fault is in table 1.

Make the decision table with mechanical system fault information， listing behavior fault examples as fault attributes （including fault attributes and fault effect attributes）. In fact， we can get the decision table 2 with mechanical system index and discrete data only through the simply replace of table 1， so the discretization is needn’t.

TABLE 1 Original data of mechanical system fault

（In table 1： SPL stands for sample；I stands for index；N stands for normal；OT stands for overtop；OL stands for overlow；AL stands for alarm；F1 stands for Feeding port congestion； F2 stands for Transfer machine fault； F3 stands for Threshers fault； F4 stands for Dryer fault； F5 stands for Sealing machine fault.）

TABLE 2 Mechanical system index and decision on faults

Reduce the table 2 with delete algorithm：

First of all， adopt the condition attributes reduction to remove the unnecessary information of the fault reasons. Delete the attribute ， which you can see the same undistinguished relationship between the original decision table and new decision table after deleting the first row of the original decision table （table 2）. That is to say， the only decision attributes can be certain according to the other condition attributes except . Similarly， the undistinguished relationship is still same between the original decision table and new decision table after deleting the condition attributes and ， which means the condition attribute is unnecessary and can be deleted. Then we can have the attribute reduction sets . The table 3 is the decision table after reduction.

Secondly， delete the repeated line in decision table. Obviously， there isn’t the same line in this example.

Finally， the redundant attributes of every decision rules can be deleted through calculating core value of condition attributes in decision rules. In table 3， in the first line brings into correspondence with ， while has the conflict with in the second line； and has the conflict with in the third line. So the attributes value can be reduced， but and can’t as the core value of the first rule. in the second line brings into correspondence with ， while has the conflict with in the first line， and has the conflict with in the third line. So the attributes value can be reduced， but and can’t as the core value of the second rule. Core value table of only decision rules can be analogized with the rest in turn

Compared to the original data， these rules are more meaningful， more concise and clear or policymakers. Now only three condition attributes （index） can achieve to the same diagnosis effect as the six condition attributes， and the attribute reduction rate is as high as 50%. Therefore， fault diagnosis rules after value reduction with rough set theory have completely consistent quality with the original decision table. What’s more， it is worth noting that the same fault diagnosis can be gotten through a part of the three attribute value. Thus， this processing not only reduced the costs of the mechanical system fault index collection， but reduced the number of dimensions of expressing space information， accelerated the speed of the mechanical system reasoning， and greatly increased the efficiency of the fault diagnosis.

4. CONCLUSION

This paper proves rough set theory is a kind of effective method in agricultural machinery fault diagnosis rules acquisition through application practice in a certain type of agricultural machinery fault diagnosis. Of course， rough set theory can also be used for other aspects， such as the fault diagnosis of the dynamic environment for acquiring rules.

5. Acknowledgements

Fund Project： Science and Technology Planning Project of Baoding ――Research on Fault Diagnosis Technology of Sewage Treatment Based on Rough Set and Artificial Neural Network （ANN）.

Reference

[1] Yinzhong Ye， Rifang Pan， Weisun Jiang. Inspection visit and the diagnosis of the fault diagnosis of dynamic systems [J]， Information and Control .1985， 6；27-34.

[2] Laibin Zhang， Zhaohui Wang， Xiyan Zhang， Jianchun Fan. Machinery and equipment failure diagnostic techniques and methods [M]. Beijing：Petroleum Industry Press，2007.

[3] Z.Pawlak ， Rough sets [J]. International Journal of Computer and Information Sciences， 1982，11：341-356.

[4] Z.Pawlak， Busse JG， Slowinski R， et al. Rough sets [J]. Communication of the ACM，1995，38（11）：89-95.