Vibration Diagnostics by Expert Systems that Link to Machinery Databases

THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS 345 E. 47 St., New York, N.Y. 10017 91-GT-298 ]^L The Society shall not be responsible for statements or opinions advanced in papers or in dis cussion at meetings of the Society or of its Divisions or Sections, or printed in its publications. Discussion is printed only if the paper is published in an ASME Journal. Papers are available from ASME for fifteen months after the meeting. Printed in USA. Copyright 1991 by ASME Vibration Diagnostics by Expert Systems that Link to Machinery Databases JAMES A. HEWITT Engineer RONALD F. BOSMANS Corporate Mechanical Engineer Bently Nevada Corporation Minden, Nevada 89423 ABSTRACT An Expert System is a software program that simulates the thought process of experts. The addition of an expert system to a machinery monitoring system provides a diagnostic monitoring system that is easier to use and understand. This paper describes a system that automatically analyzes the data collected by an online machinery monitoring system. It examines the system architecture required to add intelligent diagnostic capabilities to the monitoring system as well as the data necessary for adequate diagnosis. An example of the system's operation is also presented. INTRODUCTION Companies have become interested in Expert Systems because of the great potential for disseminating the knowledge that has been acquired in Machinery and Machinery Vibration Diagnostics. This expertise has already led to the development of many computer-based machinery condition monitoring and diagnostic systems. While these systems have helped to collect valuable data, they themselves do not distribute the knowledge and expertise required to diagnose problems. Attempts have been made to share this knowledge in the form of seminars and papers presented at machinery conferences. DESCRIPTION OF AN IDEAL EXPERT SYSTEM ARCHITECTURE In view of what expert systems - and software in general - can do today, certain features should be supported by the architecture of the Integrated expert system. These characteristics contribute to the reliability of the diagnosis, the ease of use, and to the maintainability and adaptability of the system. Automatic Data Acquisition The data for the expert system should be acquired automatically by an on-line condition monitoring system. The benefits of on-line automatic monitoring have already been discussed (Harker, 1989). In an on-line monitoring system, the plant data is collected and stored in the machinery information database. All of the important trend and transient data measurements that are required to obtain an accurate diagnosis are automatically captured by the on-line system and will be available for diagnosis at any time. The on-line monitoring process, however, merely captures the condition data: an Ideal expert system architecture must allow the Expert System to obtain that data directly from the database. This approach eliminates the possibility of data-entry errors while it vastly reduces the time required to complete an analysis. A session-driven expert system requires the user to answer computer prompted questions. It is not unusual for the user to manually answer over 100 questions before the system renders its diagnosis! An incorrect answer (or even a "typo") during almost any of the answers could invalidate the diagnosis and require the re-entry of all the data. Furthermore, an expert system that extracts its data directly from the machinery database relieves the expectation that the user must be a vibration expert who can correctly answer potentially subjective questions. A diagnostic expert system requires various forms of data to render its diagnosis. Before the expert system can process data to form a hypothesis, the raw data must be pre-processed into a form the expert system can use. Thus, an ideal expert system architecture must include a special module for extracting the diagnostically useful features from the data in the database. This is called information extraction and is a form of data abstraction. Flexibility in Database Interface The information extraction module further Isolates the actual expert system and its knowledge bases from any particular machinery database. While the data extraction portion of the module remains the same, the actual database interface can easily be changed to allow the expert system to process the data derived from virtually any condition monitoring system provided the database has the data from which the necessary features can be extracted. This flexibility in the architecture permits the expert system knowledge bases to concentrate solely on machinery diagnostics independent of how the vibration data is actually obtained. Ease of Modification The expert system's knowledge representations are independent of the source of the diagnostic data, therefore, knowledge bases can be easily maintained. Moreover, the expert system should provide a simple facility to allow the knowledge Presented at the International Gas Turbine and Aeroengine Congress and Exposition Orlando, FL June 3-6, 1991

bases to be edited such that refinement can be incorporated to account for characteristics of his own particular machines. In addition, the machine-specific knowledge bases can be modified to generate knowledge bases for other machines. For example, the rules for the diagnosis of turbine malfunctions will be slightly different from the rules for a generator. However, the rules for the generator can be produced by modifying a copy of the turbine knowledge base without recreating the entire knowledge base. Reference Machine and System Information Reference information required by the expert system concerning the configuration of the machine train must be obtained during installation and then stored for future access. This reference information is retrieved from files when a analysis is performed. This referencence information can be edited In the event of data entry error at installation or if a major overhaul of the machine train does change some particular aspect of the machine train configuration. Rotating machinery utilized In continuous service and classified as CRITICAL or ESSENTIAL to a particular process, typically contain permanently installed transducers and monitoring Instrumentation. Many of these monitoring system are supported with a host computer. Computer supported monitoring provides the function of data acquisition, data management, and presenting the data in various plot formats for review by plant personnel. Effective use of this data requires continuous and rigorous review by rotating machinery specialists. The quantity of data to be reviewed can overburden the machinery specialist to the point he feels he Is DROWNING IN THE DATA. The key element to effectively utilize this data base requires a conversion of data to INFORMATION. Once this process has occurred, the information can be extracted and an assessment of the mechanical integrity of a machine can be determined. The assessment of a machine for potential malfunction identification requires a machinery specialist, highly trained in diagnostic procedures and malfunction recognition. Recognition of the limited number of specialists and their limited availability, has given rise to exploring alternatives to performing machinery condition analysis. The evolution of EXPERT SYSTEM technology now provides a vehicle to perform machinery audits in a time effective and comprehensive manner. Two major assumptions/misconceptions regarding the use of an expert system must be avoided. First, the expert system does NOT replace the machinery specialist. Secondly, an expert system.annot be relied upon to detect and define all possible machine problems. It is limited by the information available and the knowledge embodied in the For these reasons a more appropriate name for this process is an Engineering Assist It will assist the machinery specialist in malfunction recognition, and provide recommendations as to possible corrective action requirements. The machinery specialist will review the conclusions and make the final decision regarding corrective action. WHAT ARE THE KEY ELEMENTS OF AN ENGINEERING ASSIST PROGRAM? D) Malfunction reports which include the Information utilized to reach a conclusion and the diagnostic procedure utilized. E) Explanatory text which describes the nature of the malfunction and the recommended corrective action. F) Executive summary report for each machine evaluated. This report Includes a summary of all malfunction conditions which were tested, and lists malfunctions found and malfunctions which were eliminated. ENGINEERING ASSIST PROGRAM ARCHITECTURE Four software modules are utilized 1) Data acquisition module: This module Is part of the computer hosted monitoring system. It provides data in the form of current values, dynamic wave form data, and trend files. 2) Information extraction module: a. This module extracts vibration data from the data acquisition module and processes this data to information for use within the engineering assist b. This module extracts process variable data for use within the engineering assist 3) Machine parameter module: This module provides information and a description of the machine to be evaluated. Information includes machine type, bearing type and design clearances, rotor geometry (number of blades or gear teeth, etc.), frequency of balance resonances, etc. 4) Engineering Assist Program module: This module accesses Information from the machine and parameter extraction modules and performs the assessment of current machine condition. An information flow diagram can be seen in Fig. 1 YAawwr DATA OAfA K^A9Xa Yt^ BAYS =_---_-- G'PVl M9 RX,X wzn sr^a^ (cwwrts) A) On-line data acquisition from an existing data base. B) Processing (converting) data to an INFORMATION format suitable for use by the engineering assist IKWIAIgx mut j Ib T S,SiEY -rz.a IN S1SlEu oua uawm ON D ages C) Machine condition assessment provided through the use of available information; ie. no user interface required. OAaoss 0,0 XEVwIS FIGURE l Overall Flow Of Dete Through Expert System

The engineering assist program module consists of knowledge bases and custom rule sets. Each knowledge base is focused on detecting a particular machine malfunction. Depending upon the complexity of defining a given malfunction condition, forward and backward chaining to additional knowledge bases is utilized. For each potential malfunction condition, the following Information is evaluated. These comparisons often reveal significant changes or degradation of the synchronous dynamic stiffness properties of the rotor system. An Engineering Assist program flow chart is provided in Fig. 2. A) MAGNITUDE OF THE VIBRATION - this Information is utilized to define the severity of the problem and the urgency of the corrective action. B) FREQUENCY OF THE VIBRATION - this information is utilized as an aid in classifying the category of the malfunction; le subsynchronous versus synchronous versus supersynchronous. C) FORM OF THE VIBRATION - this information is utilized to determine the shape of the dynamic motion path of the shaft centerline during a cycle of vibration. This expressed as an orbit form (circular,elliptical,highly elliptical, etc). D) AMPLITUDE/PHASE ANGLE - this information is utilized to define the modal shape or distribution of the vibration at various lateral locations along the axis of rotation. It is further utilized to determine the source location of a malfunction, such as fluid induced instabilities. E) ROTOR CENTERLINE POSITION - this Information is utilized to determine the location of the rotor centerline relative to the geometric center of the bearing. This information is presented as total shaft movement from at rest in the bottom of the bearing to the current on-line position. The shaft attitude angle within the bearing is also calculated from this measurement. In addition, rotor position as an eccentricity ratio (rotor position/bearing clearance)is calculated. F) PROCESS VARIABLE INFORMATION - this information is used In conjunction with the vibration data. It provides cross correlation between the observed vibration behavior and the current operational mode of the machine. These variables include load, flow rate, pressures, temperatures, etc. G) MACHINE GEOMETRY INFORMATION - this Information is utilized to assess malfunction Identification to known geometries of the rotor system, such as number of blades or vanes, frequency of balance resonances, etc. H) TREND FILE INFORMATION - this information is utilized to examine the behavior of the machine over a period of time. This time history reveals the rate of change of vibration or process parameters. It answers questions such as 'Was the change gradual or a step change?"; "Was the behavior linear or nonlinear?" The minimum time frame for a trends analysis should be four weeks. I) MACHINE HISTORY FILE - this Information is utilized to answer the question "What's different or what has changed in the dynamic response of the machine?" Reference files of transient data acquired during previous startups or shutdowns can be compared to current observed behavior. LEGEND FIGURE 2 eat-y^ x STRUCTURE OF KNOWLEDGE BASES As the information is processed through the engineering assist program, text screens appear advising the user of malfunctions which were defined and malfunctions which were eliminated., The user is then instructed to request additional information regarding the defined malfunction and possible corrective action recommendations. Reviewing the process utilized by the program to determine the presence of a malfunction provides training to the user in diagnostic procedures for vibration analysis. In addition, the information regarding the malfunction includes a description of the nature of the problem and the forcing functions acting on the rotor which produced the observed vibrations. In this sense the engineering assist program becomes a teaching tool. The user becomes acquainted with commonly used terms to describe various malfunctions and the underlying engineering principles. This aids in the understanding of the nature of the problem. This training tool will provide insight into the correction recommendations and will lead toward avoidance of the malfunction in future operation of the machine. Many expert systems currently In use today rely upon the frequency of vibration as the primary indicator of the type of machine malfunction. Frequency of the vibration is an important piece of information. However, if relied upon as the prime indicator of a malfunction, it can be misleading. Additional information must be acquired in order to produce an accurate diagnosis. An example of an erroneous conclusion can be found by examining Fig's 3-5. Fig.3 is a spectrum display of a vibration signal from a compressor operating at 6500 rpm. The spectrum display indicates the major frequency component of vibration is occurring at 3250 cpm. This places the class of the problem as subsynchronous vibration. Fluid induced instabilities (in general) occur at frequency ratios of 30-50% of rotor speed. If this were the only piece of information utilized, a malfunction diagnosis of oil whirl or whip would determined. However,in order for this to be true, additional information must be evaluated. For a fluid induced instability to be present, the form of the vibration must indicate a circular orbit(see Fig.4), the centerline motion must be in forward precession, and the rotor centerline position must indicate a low eccentricity ratio. The actual form of vibration (for this example) can be seen in Fig.5. The orbit is clearly non-circular, and centerline motion has both forward and reverse precession components. This additional

information eliminates a fluid induced instability as the malfunction diagnosis. Text Block 1 is an example of a malfunction text, from the engineering assist program, evaluating a power generation steam turbine. This text provides information to the user as to the data which was utilized to determine the malfunction condition of mass unbalance. In addition it describes the operating condition of the machine when the analysis was performed and other conditions which could contribute to the observed vibrations. Text of this nature provides a comprehensive evaluation of the condition of the machine and a recommended course of action. Text blocks 2 & 3 are examples of summary reports directed toward the machnery specialist and plant manager. TEXT 1 Frequency - cpm Machine Speed = 6,500 rpm M n _ H CCW Data As Observed by an Oscilloscope HLI L HNçJ 0.500 mils PP / dc Machine Speed = 6,500 rpm Fig. 4 CCW Data As Observed by an Oscilloscope rn vr/ H INFORMATION Unbalance was determined by examining the database and discovering the following facts. Overall vibration for VERTICAL = 2.6 mils pp Overall vibration for HORIZONTAL = 4.84 mils pp Slow roll compensated Synchronous (1X) vibrations: VERTICAL = 2.55 mils pp HORIZONTAL = 4.1 mils pp The shape of the orbit is HIGHLY ELLIPTICAL The Precession of the orbit is FORWARD. The data was also tested to determine the magnitude of the slow roll vibration versus compensated 1X vibration at operating speed. Slow roll vibration was found to be less than 20% of the compensated 1X vibration. The current load on the system Is 300MW. This is 75.00 % of normal load. Some turbines are sensitive to thermal variations, or steam inlet/flow variations which may effect the balance state of the unit. If variations reduce as the unit returns to normal power generating operation, balancing may not be required. An increase in the 1X vibration can be induced by an increase in the synchronous forces, such as mass unbalance. It can also be the result of a degradation of the synchronous dynamic stiffness of the rotor system. This reduction of stiffness may be caused by increased or excessive radial bearing clearances, or rotor centerline position at a very low eccentricity ratio. Your current eccentricity ratio is 0.20. Eccentricity ratios less than 0.3 indicate that the effective bearing stiffness has been significantly reduced. This will result in an increase of 1X vibration, with no actual increase in mass unbalance force or change of the balance state of the rotor. Your current D.C. probe gap voltages are -9.32 and -9.51. Upon shutdown of the unit, the final gap voltages should be measure with the rotor at rest. The difference or change in gap voltages Indicates the rotor position change from the on-line position to the position at rest. This total change Is controlled by the available bearing clearance. The design clearance for this bearing is 10 mils. If probe gap information indicates movement greater than design clearances, the bearings should be inspected for wear or damage. TEXT 2 ENGINEERING REPORT 0.500 mils pp / diu Machine Speed = 6,500 rpm The HP Turbine was examined for malfunctions Fig. 5 Survey Date 02 November 1990 Survey Time 2:28:42 pm

BEARING NUMBER 1 The condition of mass unbalance was defined for this machine. The set point limit for this condition is 3.0 mils pp. The current synchronous (1 X) vibration values are as follows: BRG#1 HP Vert = 2.55 mils pp @ 143 degrees BRG#1 HP Horiz = 4.10 mils pp @ 111 degrees This represents a 36% violation of the set point limit. The phase angle relationship (BRG1 vs BRG2) indicates the unbalance force distribution is a first mode unbalance problem. The direct influence vector for this mode is 0.5 mils/in-oz @0 deg. Corrective balance weight size must be determined based upon the weight-add radius available. Additional balancing runs (if required) and corrective weights should be determined through the use of a multiplane balancing TEXT 3 EXECUTIVE REPORT The HP Turbine has been examined for malfunction conditions Survey Date 02 November 1990 Survey Time 2:28:42 PM This turbine currently has a mass unbalance problem. It is operating at vibration levels of 36% greater than the specified limit. The turbine can be operated for a limited period of time under current conditions. However, continued changes or degradation of the balance state will require an immediate shutdown of the unit. An outage of this machine should be planned for the near future. Plant engineering has been advised of this problem and have been given recommendations regarding the initial corrective weight placements. SUMMARY Expert systems must be viewed as engineering assist programs for the machinery specialist. They must provide a comprehensive audit of the condition of the machine and provide actionable information in a format which can be understood by the machinery specialist. Failing in any of these tasks can produce erroneous conclusions and recommendations, which quickly leads to a lack of confidence in the integrity of the REFERENCES Bently, D.E. and Muszynska, A., "Early Detection of Shaft Cracks on Fluid-Handling Machines", Proc. Intnl. Symp. on Fluid Machinery Troubleshooting, ASME, December 1986. Harker, R.G. and Handelin, G.W., "Enhanced On-Line Machinery Condition Monitoring Through Automated Start-Up/Shutdown Vibration Data Acquisition", ASME Paper No. 90-GT-272.