Abstract: This paper studies a fault diagnosis system for CNC lathes. The system software is written in VC++ 6.0. It summarizes the experience of domain experts, establishes a fault tree, and constructs a knowledge base using production rules. Based on the characteristics of fault diagnosis, the system adopts an imprecise reasoning and finite depth search strategy, enabling it to diagnose more than 80 faults of CNC lathes. 1 Introduction Fault diagnosis and maintenance of CNC machine tools has always been a difficult problem for operators and maintenance personnel. Since the safety and reliability of CNC machine tools directly and significantly impact the efficiency of production units, condition monitoring of CNC machine tools for fault diagnosis and maintenance is crucial. The fault diagnosis technology for CNC lathes is based on absorbing and applying achievements from other fields, combined with the fault characteristics and mechanisms of CNC lathes themselves. By establishing a fault tree, the various components of the lathe are feasiblely divided. The fault mechanisms, causes, manifestations, and troubleshooting strategies of each component are summarized in detail, establishing a corresponding knowledge base and rule base. Appropriate knowledge representation forms are adopted according to the needs of the lathe to achieve higher diagnostic results. Because the fault rules of equipment have a certain degree of ambiguity in practice, i.e., uncertainty, this paper introduces the rule confidence approach to solve this problem, making the fault diagnosis expert system more consistent with the actual situation of the system. Using Visual C++ 6.0 as the development tool and combining database technology, a CNC lathe fault diagnosis expert system software was developed, and its feasibility was verified through field testing. 2. Structure of the Expert System The CNC lathe fault diagnosis expert system mainly consists of a knowledge base, database, inference engine, knowledge acquisition part, interpretation part, and human-computer interface, etc., and its internal structure is shown in Figure 1. [align=center] Figure 1 Structure diagram of CNC lathe fault diagnosis expert system[/align] 2.1 Knowledge Representation In an expert system, knowledge representation refers to how to conveniently store useful knowledge in the computer and facilitate the management and processing of this knowledge. The appropriateness of the representation scheme is very important for solving the problem. In this expert system, the machine learning method of the knowledge base is machine learning, that is, these rules are directly input into the fault diagnosis expert system without any processing. This system's knowledge base is constructed specifically for CNC lathe faults. Through detailed analysis of the working principles and specific structures of CNC lathes, diagnostic knowledge was obtained. Each fault state is divided into multiple records according to its cause. The knowledge base file format is as follows: The fault diagnosis system's knowledge base mainly consists of five parts: diagnostic fault tree, diagnostic rule base, fault causes, detection prompts, and fault countermeasures, comprising three files: fault tree file, rule base file, and fault countermeasure file. The diagnostic fault tree file mainly contains the fault types of the diagnosed object and explanations of the possible causes of these faults; the fault rule base file mainly contains the rules used in the fault diagnosis reasoning process; the fault countermeasure file mainly includes the corresponding elimination countermeasures for each rule and the detection process prompts for each condition item in the rule. For example: Fault tree file: Component name + "Fault tree" For example: Spindle box part: [Spindle malfunction] Number of sub-nodes = 8 ... Sub-node 5 = There are chaotic ripples on the surface of the outer circumference of the precision-machined circle #14 #15 #16 #17 # Fault cause 5 = The spindle, bearing and other transmission components are in severe friction, the raceway of the spindle rolling bearing is worn, the axial clearance of the spindle is too large, there is a gap between the outer ring of the spindle rolling bearing and the headstock shaft hole ... Rule base file: Component name + "Rule base" For example: [Rule 14] Precursor = Spindle malfunction Subsequent = There are chaotic ripples on the surface of the outer circumference of the precision-machined circle Condition information = 1 0.9 Condition 1 = Is the raceway of the spindle rolling bearing worn? #0.9 Detection prompt 1 = Please check the bearing supporting the spindle Fault countermeasure = If there is wear, replace it immediately 2.2 Knowledge management Knowledge base management mainly includes functional sub-modules such as browsing the knowledge base, creating the knowledge base, expanding the knowledge base, adding rules, modifying rules and deleting rules. The following explanation, using rule addition as an example, illustrates how this lathe fault diagnosis expert system manages its knowledge base. Adding a fault tree node involves not only operations on the fault tree file but also on the rule base file, writing new rules into it. Similarly, adding and deleting rules in the rule base file requires modifying the corresponding content in the fault tree file. The knowledge base editing function provides users with tools for adding, deleting, and modifying rules. To ensure that only authorized technical personnel can edit the knowledge base, a password is set for the editing function. First, click the "Fault Tree Management" button (using the shortcut F2). Then, perform the corresponding fault tree operation. After clicking the button, you will enter the fault tree management interface. To operate on the fault tree, click the corresponding button and press the "Add Fault Tree" button (shortcut key F2). Additionally, select the corresponding item from the "Fault Tree" section to add to the fault location in the tree. For example, if you select the fault "Spindle malfunction," once you enter content in the blank box at the "Child Node" location, the "Add Rule" and "Confirm Input" buttons will activate. Use the Tab key to shift the input focus to the "Add Rule" button and press Enter to enter the Add Rule interface. Enter the required items; if "Sensor Location" is needed, select it from the drop-down list. After completing the rule addition to your liking, press the "Confirm Input" button (as shown in the image above) to end the operation at this level; alternatively, you can click the "Cancel Input" button to abandon the addition operation. During the operation, you can use the Tab key to shift the focus. After returning, click the "Confirm Operation" button to add the fault tree operation. If you do not need to add the fault tree operation, you can click the "Cancel Exit" button to abandon the operation; you can also use the Esc key to abandon the operation. 2.3 Design of the inference engine Based on the characteristics of fault diagnosis of this CNC lathe, the design of the inference engine mainly considers the following points. (1) Inference direction This system adopts the forward inference control method. Users can find the cause of the fault based on the fault that occurs in the lathe. By matching the keywords entered by the user with the information in the diagnostic knowledge base, the corresponding record is obtained through inference, and then the cause of the fault and the corresponding handling method are displayed. The inference process is to first start the system, load the knowledge base to be used. The knowledge base called in this paper is the fault diagnosis knowledge base of CNC lathe. Then the program enters the initial interface. In the initial interface, select the menu "Diagnosis System". The system enters the diagnosis interface and prompts the user to input the fault phenomenon. After the user finishes inputting, click the "Start Diagnosis" button. The system starts inference. After the inference is completed, the user can see the corresponding fault cause and solution proposed by the system in the output result interface. (2) Search method The finite depth first search method is adopted. It does not only require that the unsearched child nodes always search to the "depth", but gives a depth limit coefficient. When the depth difference of the search reaches the depth limit coefficient, it backtracks appropriately, which can reduce the workload of the search. (3) Uncertainty fault phenomena are obtained from observation and generally cannot be represented in a definite way. It is necessary to attach a confidence level [0, 1] to each rule clause. (4) Design of the interpreter This part of the functional module is responsible for answering various questions raised by users, including questions related to system reasoning and questions about the system itself that are not related to reasoning. It can provide necessary clear explanations of reasoning routes and meanings, providing users with a convenient means to understand the reasoning process and system maintenance. It is the main module for realizing system transparency, and it also facilitates expert systems and knowledge engineers to discover errors in the knowledge base in a timely manner. During the reasoning process, the rules and variable transformations based on each step of reasoning are recorded in the dynamic database in chronological order, forming a continuously extending "chain". Once the outside world requests an explanation, the interpreter will organize this chain appropriately and display the reasoning process to the user. (5) Interface The interface consists of some menu commands. It mainly consists of: system start command, system diagnosis command, knowledge management command, parameter setting command, system exit command, explanation command, input and output command, etc. 3. System Implementation This system utilizes Visual C++ 6.0 to develop the user interface for a CNC lathe fault diagnosis expert system on Windows. The interface is displayed upon system startup. Users can run the system following menu prompts and select operations from the toolbar or menu during system operation. The knowledge base is currently being expanded. Due to the ease of adding and deleting diagnostic knowledge, this system essentially functions as the skeleton of an expert system, continuously enriched and improved. 4. System Functions Diagnoses various faults in CNC lathes. It includes an explanation function (answering "why?") and machine learning capabilities. During diagnosis, users simply select the relevant fault phenomena and test data, and the system draws a conclusion. 5. Conclusion This system achieves the diagnosis of various faults in CNC lathes, reaching a level comparable to that of a general expert. Lathe operators and maintenance workers can easily master and use it. Trial testing has shown that the diagnostic conclusions are generally accurate.