Knowledge-Based Diagnosis System of Machine ?· Knowledge-Based Diagnosis System of Machine Tools Mujin…

  • Published on
    24-Jul-2018

  • View
    212

  • Download
    0

Embed Size (px)

Transcript

<ul><li><p>Knowledge-Based Diagnosis System of Machine Tools</p><p>Mujin Kang*, Dong-Kyu Seo**</p><p>* Sung Kyun Kwan University** LG-EDS</p><p>Abstract</p><p>This paper outlines the development of a knowledge-based expert system for diagnosing machine toolfailures. The complexity of machine tool failures andthe trouble shooting procedures are deployed. Thediagnostic knowledge is acquired from analysis of thetroubleshooting and repair reports as well as frominterviews with diagnosis engineers. The knowledgebase is built using the object oriented modeling. Theinference mechanism mainly based on backwardreasoning, the flow of diagnosis process guided by theexpert system, and the user interaction are explainedwith an example. A test run of the developed prototypesystem, developed with a commercial expert systemshell, shows the feasibility of local problem solvingwithout help of a service person sent from the machinetool maker.</p><p>Keywords: expert system, knowledge base, objectmodel, machine tools diagnosis</p><p>1 Introduction</p><p>In an environment of todays ever increasingcompetition, a manufacturing company needs to makefull use of its manufacturing facilities and thus shortenthe manufacturing lead time in order to meet thecustomer due dates. It requires not only an effectivemanagement of operations but also maintaining theoptimal condition of the manufacturing system. In caseof machine failures the company should be able toisolate the causes as rapidly as possible and to takeappropriate corrective actions so that the MTTR (meantime to repair) be kept minimal.</p><p>Machine tools for processing materials are one of themost important elements in a manufacturing system.Troubleshooting of modern machine tools equippedwith sophisticated electronic as well as mechanicalcomponents is so difficult that it requires a lot ofexperience and knowledge, and thus generally dependson the diagnosing persons from machine tool makers.</p><p>The character of machine tool, being capitalinvestment goods unlike consumer products such ashousehold electric appliances, makes it expensive andalmost impossible for a machine tool maker to run avast service network enough to support their customersquickly. An unmanned diagnosis system to which theusers can have access at all times and get help in caseof a trouble occurrence could be an efficientalternative[1].Diagnosis in its general sense includes monitoring andsupervision of machine operation as well as finding outthe causes of a trouble. The objective of the former isto recognize any possible abnormal condition early andprevent machine failures opportunely by keeping trackof the processes. Many previous works have been doneon this problem, and remarkable achievements havebeen made in the last years. They can be classified aseither condition-based or model-based[2][3]. On theother hand, efforts to analyze causes of a trouble andsuggest solutions are rarely found.This paper addresses the diagnosis of machine tools inits narrow sense focusing on the fault systematics andthe relationship between symptoms and causes.Conventionally, machine diagnosis has been madeusing fault tree analysis or decision table technique,where the troubleshooting paths must have beenpredetermined. The advantage of these approaches isthe simplicity in use and the visibility of the problemcomplex, while they are not easy to maintain. Anyway,they provide the basis for more advanced ones likerule-based or case-based reasoning, which seem to bepromising in solving the cause and effect relationshiptype problems[4][5][6]. Some of the related researchessuggest a new possibility of applying the diagnosissystem in a server-client environment, i.e. remotediagnosis. It performs the diagnosis tasks based on theinformation supplied through internet from the user ata remote site, and sends the results back [1][7][8][9].The premise is, of course, that an efficient diagnosisserver being able to identify the causes of a failure isprovided.This paper describes a knowledge-based expert systemfor diagnosing machine tool failures, which isdeveloped for a CNC machine tool maker.</p></li><li><p>2 Diagnosis of machine tool failures</p><p>Failures occurring in machine tools are so diverse thateven classification is not easy. However, they can beroughly grouped into four categories no operation,malfunction, inaccurate machining, and the others. Nooperation means that a component of the machine toolis not acting, while malfunction means an abnormalmovement of a machine element. Inaccurate machiningmeans inferiority such as bad surface roughness oroccurrence of unwanted taper. The frequencydistribution for each group is shown in Figure 1, whichhas been taken from the service statistics of a machinetool company. It indicates that the most frequenttrouble is no movement of a machine element such asautomatic tool changer, spindle, or hydraulic chuck.</p><p>Figure 1. Classification and frequency of troubles.</p><p>Diagnosis is a reasoning process to analyze the causeand effect relationship beginning from the symptom,and to find out the origin of the failure. Difficulty inidentifying the causes has been attributed to variousfactors like system complexity, inaccurate information,and lack of adequate troubleshooting tools. In order toform the basis for rational troubleshooting, the possiblecauses for each trouble should be listed up first, andthen the logical failure-cause relationships are to beexplored stepwise into more detail. For that purpose itis essential to understand the diagnosing experts wayof thinking. Figure 2 shows an example of diagnosisworkflow for inaccurate machining case. At theoccurrence of a taper on axis-symmetric cylindricalsurface, NC program, machining parameters, diagnosisnumber(DGN) are sequentially checked until theultimate cause is localized. If no cause can be found inthe control unit, the backlash of lead screw, thealignment of table, spindle, tail stock, and tool post,etc. are examined in that order. Thus, the diagnosticknowledge has been acquired from analysis of thetroubleshooting and repair reports as well as frominterviews with diagnosis engineers. An excerpt of theknowledge base built is shown in Figure 3</p><p>Figure 2. A typical example of troubleshootingworkflow.</p><p>3 Object-oriented representation of thediagnostic knowledge</p><p>The object-oriented approach is a good match to themachine tool diagnosis problem for a couple ofreasons. Firstly, the physical configuration of amachine tool, usually represented in a BOM (Bill ofMaterials), shows a hierarchical structure, which givesa natural object-class hierarchy. According to theproduct structure, the cause of a failure is propagatedfrom an original source to a higher-level assembly.Secondly, the diagnostic logic also has a hierarchicalstructure reflecting the functional interaction of thecomponents. The diagnosis can be regarded a reversetracking process of the failure propagation. As shownin Figure 3, the diagnostic knowledge can berepresented in an object network, which is natural forthe class inheritance features. As the highest-levelobject the class Failure_Symptom is generated. It hassub-classes such as Malfunction, No_Operation, andInaccurate_Machining. The properties like Taper,Stripes, Roughness, etc. are again assigned toInaccurate_Machining.</p></li><li><p>Figure 3. An excerpt from the diagnostic knowledge base.</p><p>Figure 4. Reasoning mechanism in backward chaining.</p><p>The diagnostic knowledge is represented in the form ofproduction rules using a commercial expert systemshell. A rule is composed of conditions, hypothesis,actions, and optional alternative actions. Thehypothesis is a kind of object representing theconclusion of one or more rules. The class structureitself acts as a pointer to a set of objects, with data heldby the structure formed by associating a property witheach object. This structure is represented by the formclass_name.property_name and is known as a slot.Classes thereby provide a way to search through listsof objects in order to identify which objects meet aspecific condition, which is called pattern matching. Italso enables the maintenance of failure occurrencestatistics by updating the assigned property for rulefrequency in accordance with the pattern matchingresults.In the backward chaining which is predominantly usedfor diagnosis problem, reasoning proceeds from a goaltrying to match the known facts. It is based on the</p><p>evaluation of the hypotheses. The backward reasoningmechanism is illustrated in Figure 4. Let the goal behypothesis hypo1. To evaluate hypo1 of the rule withunevaluated condition, hypo2 which itself is acondition of the rule must be verified to be true. Thesystem tries to find one or more rules with hypo2 as thehypothesis. Once a rule with hypo2 as its hypothesis isfound, the conditions in the rule leading to hypo2 areevaluated before finishing the evaluation of theconditions which caused the backward chaining tooccur.In the system developed, some more inference searchmechanisms like Subgoal forward, Semantic gates, andForward actions effects are used (Figure 5). Subgoalforward is a consequence of investigating subgoals asopposed to a terminal hypothesis. It consists of notonly exploring the backward chaining associated withthe subgoal, but immediately thereafter placing on theagenda the hypotheses of the rules in which subgoal isinvolved in a data. An important feature is that subgoalforward events are queued and evaluated as soon as thecurrent rule has been completely evaluated.Semantic gates, or simply Gates, are structure-basedinference mechanism which accounts for theopportunistic insertion of hypotheses on the agenda.Gates are generated during the evaluation of theconditions list of rules. When any new slot is evaluatedin the conditions of a rule, the inference engine checksto see if any other rules also have this slot in theirconditions. If any target rules have this slot in theirconditions, the inference engine tests the value ofparticular conditions which include this value. If thecondition is TRUE, the associated hypothesis is put onthe agenda. In other case, the hypothesis will remain</p></li><li><p>UNKNOWN rather than being evaluated to FALSE.While Gates pre-evaluate the target condition and onlyplace the hypothesis on the agenda if the condition isTRUE, Forward actions effects are non-selective. Thismeans they place on the agenda any hypothesis whoseconditions have a slot that has been affected by theactions list. Thus it may queue hypotheses even thoughthe condition involving the modified slot might beFALSE.</p><p>Figure 5. Various inference mechanisms.</p><p>Let us consider the case of inaccurate machining. Theinference engine searches the rules in order to prove ifthe slot value of Hypo_Failure is true. The conditioncan either be given by the user or be determined by theoutcome of firing other rules. In this example, it isassumed that the symptom Inaccurate_Machining isselected. Objects in the class Failure_Symptom allhave the property Symptom. The system evaluates thecondition and identifies all the slots ofFailure_Symptom.Symptom with the valueInaccurate_Machining. As a result, a list of objects foruse by the right hand side actions is generated. If thetype of inaccurate machining is taper occurrence,which is to be given by the user, the value Taper isassigned to Inaccurate_Machining.Symptom slot.Then, the related rules are fired according to thesequence described in Figure 2 to get the value ofInaccurate_Machining.Taper slot. An example of thefinal reasoning shows the value assignmentInaccurate_Machining in Failure_Symptom.Symptom,i.e. Symptom slot of the Failure_Symptom class, Taperin Inaccurate_Machining.Symptom, Headstock_Centerin Inaccurate_Machining.Taper.</p><p>4 Knowledge-based diagnosis system ofmachine tools</p><p>Using the knowledge base built based on the data andknow-how of a CNC machine tool manufacturer, aninitial implementation of the diagnosis expert system</p><p>has been made. The main components of the systemare knowledge maintenance module, knowledge baseuploading module, diagnosis module, and reportingmodule. The diagnosis module, core of the system,provides the functions like symptom entry, knowledgebase search and rule firing, query and input processing,and corrective action processing.</p><p>Figure 6. Screenshot of system main menu.</p><p>From the initial GUI(Graphical User Interface) asshown in Figure 6, users can select the knowledgebase, start diagnosis, modify knowledge base, or retryreasoning. The knowledge base is uploaded accordingto the selected machine tool type and the diagnosismethod symptom-based or alarm-based. Diagnosisprocess begins with the selection of the symptomgroup, for instance inaccurate machining. With thisgiven hypothesis the system performs the inference tofind the original cause. When the slot values during thereasoning cannot be obtained by the knowledge basesearch, they must be input from the user, guided by theexpert system. The question handler in charge ofprocessing questions and answers provides a user-friendly interaction environment. Through the help ofgraphical support and comprehensible questionnaires,</p><p>Figure 7. Question handler window for inaccuratemachining.</p></li><li><p>the user can input the meaningful value withoutdifficulty. For instance, a symptom can be correctlyselected from the possible defects as displayed inFigure 7. In a similar way, the object properties aredetermined in order.Considering again the example shown in Figure 2, letus suppose that the cause of the trouble taperoccurrence is backlash in the lead screw. Figure 8shows the question handler indicating the necessarywork instructions for backlash check. A typicalexample of the final reasoning result looks like Figure9, which shows the necessary corrective actionstogether with question and answer history.</p><p>Figure 8. Question handler indicating the actions forbacklash check.</p><p>Figure 9. Example of a final diagnosis result showingthe corrective actions and Q &amp; A log.</p><p>5 Concluding remarks</p><p>A knowledge-based expert system for diagnosingmachine tool failures has been developed. Thediagnostic knowledge has been acquired from theanalysis of troubleshooting and repair reports as wellas from the interviews with diagnosis engineers. Theobject-oriented approach is used to implement the</p><p>knowledge base using a commercial expert systemshell. The inference mechanisms such as backwardreasoning, hypothesis forward, semantic gates, andforward action effects are used for reasoning.Experiences obtained from test runs for a CNC latheseem quite promising. Provided a more extensive andreliable knowledge base is implemented, an unmannedsupport of the machine tool users in troubleshootingwill be possible.</p><p>Acknowledgements</p><p>The authors acknowledge the support of HyundaiPrecision Engineering and the cooperation ofHwacheon Machine Tools Co.</p><p>References</p><p>1. Kang,D.C., Kang,M., A study on the internet-basedremote diagnosis system for machine tool failures,Journal of Korean Society of PrecisionEngineering, Vol.16, No.9, pp.75-81, 1999 (inKorean).</p><p>2. Westkaemper,E., Westerbusch,R., The diagnosisa...</p></li></ul>

Recommended

View more >