Compared to even just a few years ago,

Transcription

1 TPM GETS SMART Prognostics and health management algorithms help Total Predictive Maintenance step up its game Jay Lee, FSME Director Edzel Lapira Researcher Intelligent Maintenance Systems Center University of Cincinnati Compared to even just a few years ago, the modern manufacturing environment necessitates a more elevated degree of plant and equipment reliability and availability that leverages state-of-the art technology that can convert data to usable information. The current manufacturing landscape has drastically changed and is forcing companies to reassess Total Productive Maintenance (TPM) and how to improve the gains beyond what it can already provide. The introduction of TPM several decades ago spurred a culture change within different manufacturing companies in the metal, automotive, chemical, rubber, and food industries, among others. The goals of maximized overall equipment effectiveness (OEE) and improved reliability and maintainability have compelled and sustained organizations to adhere to the TPM philosophy. Success stories from medium and largesize companies have inspired unbelievers to embrace TPM. The use of a strict operator-centric maintenance program, coupled with intensive data collection and data management, has helped many companies, in varying degrees, to reduce waste, defects, and downtime. But today s market panorama is painted in stark contrast compared to the prevailing conditions during TPM s inception. The product manufacturing life cycle, from design to production, has become alarmingly fast with new model revisions being introduced to the market at short intervals. This occurs even with significantly increased market competition in many product segments. These conditions have influenced manufacturing facilities to become flexible and agile while at the same time being operationally lean. These characteristics have become important even as many companies are still reeling from the effects of the global economic recession that engulfed the world in Production demands, in most industries, have not, so far, reached or surpassed their pre-recession levels. In addition, the complexity of most production equipment has grown exponentially with the advance of instrumentation and sensing. The multitude and variety of signals that are being collected from these machines may confuse even a skilled operator, which may hinder making objective decisions for day-to-day operations. In the original context of TPM, the operator is the first line of defense, having been enriched with sufficient technical training and knowledge, to perform primary machine maintenance. The operator is also responsible for performing data collection of production summary statistics (such as number 70 June 2011

2 Manufacturing Maintenance of parts produced, scrap parts, rework, among others) and general machine states (such as number of hours in production, repair, waiting for raw materials, among others). Together with predefined time factors (machine setup, tooling, loading, actual production, metrology for sampled parts, unloading), the OEE status is computed. TPM focuses the organization to review its operations to look for opportunities to reduce and/or eliminate waste, defects, and downtime. The reliability aspect is addressed by recording and analyzing maintenance events using details such as failed component, fault mode, number of hours to repair, type of repair accorded, and the like. By using statistical process control (SPC) and reliability-centered maintenance (RCM) tools, different metrics are evaluated that allow for a planned equipment maintenance (or preventive maintenance, PM) program. In the last decade, the adoption of enterprise resource planning (ERP) systems has allowed the automation of data collection from operator computer stations and even from machine controllers. Data connectivity standards have also been proposed to achieve a uniform data format across different machine controller manufacturers. Plant maintenance database programs or computerized maintenance management systems are also available to aggregate the collected data. The traditional implementation of TPM, as is true of even its recent evolution, can be characterized as an intensive and exhaustive data collection and data management exercise. On the other hand, smart TPM aims to utilize advanced tools and methodologies that can aid in generating metrics that reflect the state of manufacturing operation performance. Smart TPM does not replace the traditional implementation; rather, it is a natural extension and progression in order to truly achieve the benefits of the TPM philosophy. Traditional TPM has laid out the foundations for a systematic extraction of data from different facets of the facility including machine, operator, and maintenance activities. It has effectively bridged the gap between the machine and data levels. Smart TPM then converts the recorded data into health information using sophisticated prognostics and health management System-level view illustrates how Smart TPM is a natural extension and progression of traditional TPM and is meant to achieve the full benefits of the TPM philosophy. June

3 Manufacturing Maintenance (PHM) algorithms. PHM has become very popular in different application domains but has not been adopted by many companies. PHM is used to analyze or digest machine data in order to achieve an objective health assessment. A more detailed discussion of PHM is given in the succeeding section. Visualization tools such as degradation curves, fault maps, multicomponent health, and risk charts are generated to provide specific information. An operator and production supervisor may be interested to know whether a machine is fit for operation, i.e., can it run a production lot with the desired throughput rate and expected part tolerances. A maintenance technician will gain valuable insights from a multicomponent radar chart to determine which component is the most likely root cause of a fault or degradation. In addition, a fault map can aid in troubleshooting the source and severity of the failure. A maintenance department supervisor can look at a risk chart to make informed decisions in prioritizing maintenance activities with a depleted crew. At an enterprise level, the health information can then be used to generate a more reliable and objective network OEE that is based on the true condition of the plant machines. It can also be used for an optimized just-in-time maintenance program instead of a very expensive, and at times unnecessary, PM implementation. Finally, historical health information can be utilized for closed-loop manufacturing to product design improvement. Although the current literature may offer various definitions, PHM is the judicious use of the different maintenance strategies (reactive, preventive, condition-based, and predictive) when and where appropriate. Criticality analysis tools and methods can be utilized to match a particular task to a maintenance diagram. Reactive maintenance is the traditional failand-fix method that can be used for low-impact (frequency of occurrence, downtime, cost, etc.) repair tasks. Preventive maintenance can be applied in situations wherein it is difficult to achieve a consistent and reliable degradation model and if the prediction time horizon does not permit sufficient time for any meaningful machine conditioning prior to an impending fault event. Condition-based maintenance estimates the current health of the machine to gain insight into whether the equipment is mission-reliable or able to finish a part without breaking down or produce a part within the desired quality tolerances. Finally, predictive maintenance with prognostics goes a step better by automating the process of failure prediction. It does so with sophisticated algorithms that can predict, with some degree of confidence, component degradation. The Watchdog Agent is a toolbox composed of different analysis and machine learning algorithms June 2011

4 The main advantage of prognostics is its ability to provide actionable information regarding when to perform maintenance operations. As a consequence, maintenance and repair can be carried out while not disturbing the production schedule and indirectly, reducing unnecessary maintenance by extending the component or system life. For just over a decade now, the Intelligent Maintenance Systems Center (IMS) has been at the forefront of researching predictive techniques and embedded computing technologies for machine prognostics and health management that can be used to aid different manufacturing companies in their smart TPM implementation. IMS has developed a toolbox of analysis and machine-learning algorithms for predicting the degradation of components and machines. The Watchdog Agent is able to convert declining critical performance data into health features, and quantitatively assess their confidence value to further predict trends so that proactive actions can be taken before potential failures occur. The tools can be classified into four categories: signal processing and feature extraction, health assessment, performance prediction, and health diagnosis. What follows are two case studies that briefly describe how tools from the Watchdog Agent were used to aid in smart TPM. Case Study 1: Torque Monitoring of Industrial Robots This particular application deals with the health monitoring of a network of six degrees of freedom industrial robots used in an automotive manufacturing environment. The industrial robots are visible to the company s IT infrastructure via Ethernet connection. The application of smart TPM is seen with the inclusion of the Watchdog Agent in the company s data repository to analyze the data and provide maintenance personnel and plant supervisors with robot health information. Currently, the diagnostic approach adopted to monitor the robot includes monitoring the root mean square (RMS) of the duty-cycle torque for each joint and providing an alarm based expand the possibilities of gauging Introducing EQUATOR the r cr Expand process control to the next level. June

5 Manufacturing Maintenance on the values exceeding manually set thresholds. The RMS is an indication of the torque requirements (holding and motion) for a given application s duty cycle. RMS torque values that exceed set threshold levels can cause servomotor overheating and reduced headstock life. As is the case for most diagnostic systems, however, such diagnostics provide a primitive indication of a fault event that may usually be sensed too late in the process. Hence, the time window to perform the necessary maintenance or corrective action is usually too tight. In fact, maintenance is usually performed after the undesired event or failure has actually occurred. An improvement, based on PHM methodology for robot torque monitoring, was implemented to transform this diagnostic-based system to a real-time health assessment of each of the robot s joint servomotors. This allowed for a more optimized repair window and improved maintenance activities to be carried out in the earlier stages of the degradation process, as opposed to the current failto-fix approach caused by costly and untimely failures. Case Study 2: Machine Tool Health Monitoring This case study describes a bearing health-monitoring task that used an out-of-process strategy, which means the bearing health will be evaluated while the machine tool is not in operation, i.e., normal production. This is a necessary characteristic chosen for the monitoring platform to remove the effects of variable factors (spindle speed, feed rate, depth of cut, part material, and other in-cutting parameters) during prognosis. Otherwise, numerous experiments (including time, resources, and effort) would have been needed to develop accurate and robust models that would take these factors into account, which can prove to be prohibitively more expensive in the long run. Instead, the procedure, done automatically, will force the machine to perform specific actions and movements at predefined operating conditions. The strategy is implemented as a Fixed-Cycle Feature Test (FCFT) through a routine NC program run on a daily basis. An integral part of the prognostics system is machine instrumentation and the data acquisition system. Sensors MANUFACTURING NOW! More Than Ever. Media Partner SOUTH-TEC is produced by SME 74 June 2011

6 are installed on the machine, a data acquisition device (DAQ) converts the analog sensor signal into digital form and data is stored in the computer for future analysis. Communication with the Fanuc controller is used to trigger data acquisition of accelerometers. The data then undergoes four procedures to complete the conversion into health information: data segmentation, feature extraction, health assessment, and remaining useful life (RUL) prediction. Selected algorithms from the IMS Watchdog Agent toolbox including the Fast Fourier Transform (FFT), Logistic Regression (LR), Statistical Pattern Recognition (SPR) and Autoregressive Moving Average (ARMA) model are used. The PHM system then provides an estimation of the RUL of the different bearings (in hours of operation) being monitored so that component repair (or replacement) can be optimally scheduled without disrupting scheduled production. ME About the IMS Center The National Science Foundation s Industry & University Cooperative Research Center (NSF I/UCRC) for Intelligent Maintenance Systems (IMS) is a multicampus NSF Center of Excellence, founded and directed by Jay Lee in The Center is a consortium of three research sites at the University of Cincinnati (lead institution), the University of Michigan, and the Missouri University of Science & Technology, as well as three international sites (Singapore, Brazil, and Spain). The Center has worked with more than 70 global companies and conducted more than 100 projects since its inception. The vision of the Center for Intelligent Maintenance Systems is to enable products and systems to achieve and sustain near-zero breakdown performance, and ultimately transform maintenance data into useful information for improved closed-loop product life cycle design and asset management. Further information on the Center may be found at Jay Lee is a Fellow of the Society of Manufacturing Engineers and Edzel Lapira is an SME member. Designs for threading, grooving, PCD, and CBN-tipped inserts Set includes boring bar, 5 inserts, 5 screws, and a torx plus key Available in coated AlTiN or uncoated sub micrograin carbide Hardened steel head features proprietary mechanical attachment... no annealing from brazing heat T H E C U T T I N G E D G E Pocket keeps chips away from machined surface Fits into SCI standard QHC coolant tool holders Inserts lapped and ground to fine finish for maximum chip flow Available in right-hand and left-hand versions Bars stocked with or without locating flat... Low profile screw keeps chips flowing We use Premium Micro grain solid carbide for longer tool life and increased feeds and speeds. Tools are stocked uncoated and ALTiN coated. Solid carbide grooving, boring and threading tools. Threadmills Port tools SCIENTIFIC CUTTING TOOLS, INC. 110 W. Easy Street / Simi Valley, CA / / [fax] info@sct-usa.com / June