A Framework for Cost-effective and Accurate Maintenance Combining CBM RCM and Data Fusion

Transcription

1 A Framework for Cost-effective and Accurate Maintenance Combining CBM RCM and Data Fusion Gang Niu Prognostics and Health Management Centre Electronic Engineering Department City University of Hong Kong Michael Pecht Prognostics and Health Management Centre Electronic Engineering Department City University of Hong Kong Abstract Cost-effectiveness and accuracy are two basic criteria for good maintenance. Reducing maintenance costs can increase net profit, while accurate maintenance actions can sustain continuous and reliable operation of equipment. As instruments and information systems tend to become cheaper and more reliable, condition-based maintenance has become an important tool for running a plant or a factory. This paper presents a novel condition-based maintenance framework that uses reliabilitycentered maintenance to save money and employs the data fusion strategy for improving condition monitoring, health assessment, and prognostics. This framework can help obtain optimal maintenance performance with good generality. Keywords - condition-based maintenance; reliability-centered maintenance; data fusion I. INTRODUCTION With the fast development of industry and a highly competitive international market, cost-effective and accurate maintenance shows increasing importance in improving plant production availability, reduce downtime cost and enhance operating reliability [1]. There are two major types of maintenance: corrective maintenance and preventive maintenance, as shown in Fig. 1. Corrective maintenance is undertaken after a breakdown or when an obvious failure has been located. However, corrective maintenance at its best should only be utilized for non-critical areas where capital costs are small, consequences of failure are slight, no safety risks are immediate, and quick failure identification and rapid failure repair are possible [2]. Preventive maintenance is carried out at predetermined intervals or according to prescribed criteria and is intended to reduce the probability of failure or the degradation of the functioning of an item (see SS-EN 13306). Preventive maintenance is divided into two types: predetermined maintenance and condition-based maintenance (CBM). Predetermined maintenance is scheduled without the occurrence of any monitoring activities. CBM, on the other hand, does not use predetermined intervals and schedules. Instead, it monitors the condition of components and systems in order to determine a dynamic preventive schedule. Figure 1. Overview of different maintenance types. For the purpose of general structure, the open system architecture for CBM organization (OSA-CBM) [3] has divided a standard CBM system into seven different layers, with the technical modules shown in Fig. 2. The core functions, corresponding to maintenance decisions among the architecture, can be summarized as condition monitoring, /09/$ IEEE 605

2 health assessment, and prognostics ranging from layer 3 to layer 5. Condition monitoring involves comparing on-line and off-line data with expected values. If necessary, it should be able to generate alerts based on preset limits. Health assessment determines if the health of the monitored component or system has degraded and exerts fault diagnostics. The primary tasks of the prognostics are calculating future health and estimating the remaining useful life (RUL). Figure 2. Open system architecture for condition-based maintenance (OSA-CBM). In reality, however, reliable and effective CBM faces some challenges. First, initiating CBM is costly. Often the cost of the instrumentation can be quite large, especially if the goal is to monitor equipment that is already installed. It is therefore important to decide whether the equipment is important enough to justify the investment. Second, while the goal of CBM is accurate maintenance, this is not always easy to achieve due to variables such as the complexity of the environment, the inner structure of the equipment, obscure failure mechanisms, etc. For cost-effective maintenance, CBM is best implemented under an advanced maintenance management mechanism. When the functions of a component and its importance need to be considered at the same time, reliability-centered maintenance (RCM) is an appropriate choice [4]. Usually, the aim of RCM is to maximize results with regard to system reliability or outage cost reduction [5]. For accurate maintenance, data fusion techniques containing signal-level fusion, feature-level fusion, and decision-level fusion are suggested [6]. Applying fusion techniques in engineering practice has been receiving increasing attention in recent years. Especially with the rapid progress of advanced sensor and signal processing technologies, fusing large amounts of mutual information has become possible. These developments are expected to bring about accurate CBM. This paper develops a novel maintenance framework that integrates data fusion strategy with traditional CBM within the framework of RCM management. Using the data fusion strategy can increase maintenance accuracy, while RCM can help to implement CBM with optimal cost benefits. A detailed introduction of the proposed framework is discussed in the next section. The basic contents of each module are also explained. II. PROPOSED CBM SYSTEM BASED ON RCM AND DATA FUSION In this section, a new CBM system is proposed, as shown in Fig. 3. This system is based on OSA-CBM architecture and is integrated with RCM and data fusion strategy. RCM is employed in order to achieve cost-effective maintenance, while data fusion technology is used for monitoring, diagnostics, and prognostics in order to improve maintenance accuracy. 606

3 Figure 3. Flowchart of proposed CBM system. A. Cost-effective Maintenance by Integrating CBM and RCM One of the main purposes of maintenance is cost saving, which shows increasing importance in both civil industry and military industry where complex equipments maintenance cost huge money. RCM can effectively organize CBM in optimizing solution to reach the aim. A.1. Structured steps of applying RCM The general steps for applying the RCM method [7] are as follows. 1) Define system functions, performance standards, and system boundary definition. A function definition is not complete unless it specifies the level of performance desired by the user. Performance can be defined in two ways: Desired performance, i.e., what the user wants the asset to do. Built-in capability, i.e., what the asset can do. 2) Determine the ways in which the system functions may fail. A failure is an unsatisfactory condition. A functional failure is defined as the inability of an asset to fulfill a function in accordance with a standard of performance. All the functional failures associated with each function should be recorded. 3) Determine the significant failure modes. A failure mode is the effect by which a failure is observed to occur [8]. The best way to show the connection between functional failures and the events that could cause them is to list the functional failures first and then record the failure modes that could cause each functional failure. 4) Assess the effects and consequences of the failures. A failure effect is what happens when a failure mode occurs. Failure consequences are the ways that failures matter. The consequences of failures are more important than their technical characteristics. To identify the failure modes and their failure effects, a failure mode and effect analysis (FMEA) is usually performed. Often FMEA becomes a failure mode effect and criticality analysis (FMECA) if criticalities or priorities are assigned to the failure mode effects [9]. 5) Identify maintenance tasks by means of a decisionlogic scheme. The decision-logic scheme helps to evaluate the maintenance requirements for each significant item in terms of the failure consequences. It selects only those tasks that will satisfy these requirements. For items where no applicable and effective task can be found, re-design is recommended if safety is involved. Otherwise, there should be no scheduled maintenance. 6) Identify of maintenance task interval. RCM focuses only on what tasks should be executed and why. The time when the tasks should be executed is derived from separate analyses that must consider and utilize combinations of company and industry experience to establish task frequencies. Based on the above steps, different types of maintenance tasks and intervals can be selected. Corrective maintenance and critical functions of failures need to perform repair immediately, which is a broken-then-fix strategy. However, for failures that are of little or no consequence to comprehensive system function, the maintenance can be deferred until a suitable time. Predetermined maintenance is best suited to an item that has visible age or wearout characteristics and one 607

4 where maintenance tasks can be made at a time that is certain to prevent a failure from occurring [10]. The scheduling can be based on the number of hours in use, the number of times the product has been used, etc. For CBM, the core part is condition monitoring, which can be performed using various approaches and utilizing different levels of technology [11], as seen in Fig. 4. In on-line (or real-time) monitoring, one continuously monitors a machine and triggers a warning alarm whenever something wrong is detected. Periodic monitoring is used due to its low cost and the fact that it provides diagnostics using filtered and/or processed data. 7) Auditing, implementation and feedback: After the review has been completed for each asset by the RCM team, senior managers with overall responsibility for the plant s equipment must evaluate the decision made by the team, a procedure called auditing. After each review is approved, the recommendations are implemented by incorporating maintenance. Figure 4. Divisions of condition monitoring. A.2. Relationship between CBM and RCM The RCM analytical approach assists the maintenance manager in identifying potential failures and supporting the selection of viable courses-of-action. RCM tools help define the optimal failure management strategies. Meanwhile, CBM is built on the foundation of the RCM methodology. CBM is not a process in itself. It is a comprehensive strategy to select, integrate, and focus a number of process improvement capabilities, thereby enabling maintenance managers and their customers to attain the desired levels of system and equipment readiness in the most cost-effective manner. CBM strategy includes a number of capabilities and initiatives, some procedural and some technical, that can enhance the basic RCM tasks. In this way, CBM enables a more effective RCM analysis. B. Accurate Maintenance by Integrating CBM and Data Fusion Techniques In industry, accurate maintenance is imperative, especially for core functional components of machinery. Integration of CBM and data fusion methods is expected to generate accurate maintenance decisions. B.1. Proposed CBM system based on data fusion In this subsection, a CBM system based on data fusion is introduced. The suggested flowchart is shown in Fig. 5. First, raw signals are collected and signal preprocessing is conducted. Then the appropriate features are calculated and extracted that give information about the state of the running machine. Next, the features that can recognize different faults clearly are selected for diagnostic analysis, while those features indicating the degradation trend of equipment health are chosen for monitoring and prognostics tasks in order to predict the remaining useful life (RUL). Then the fusion strategy is employed in the following two subsystems: one is fusion fault diagnostics subsystem, the other is fusion monitoring and prognostics subsystem. As far as health degradation is concerned, each degradation indicator has its own merits and shortcomings and is only effective for certain failures at certain stages. Therefore, fusing multiple degradation indicators would potentially provide an accurate and reliable way to monitor degradation. When the monitored index exceeds a predetermined amount, the processes of diagnostics and prognostics are trigged. In order to get accurate diagnostics results, either the feature-level or decision-level fusion strategy can be used. Moreover, in order to get a near-ideal prediction of RUL with its uncertainty interval, data fusion at the feature-level can also be employed. Finally, optimal maintenance actions are established based on the results of monitoring, diagnostics, and prognostics. A description of the fusion subsystems is given below. B.2. Fusion diagnostics subsystem The task of the classifier (diagnostics) component of a full system is to use the feature vector provided by the feature extractor to assign the object to a category (fault) [12]. The machine fault diagnostics process takes a segment of fault signals and tells which fault it represents. In the proposed fusion diagnostics subsystem shown in Fig. 6, signals are collected and feature extraction is performed. The extracted features should be the ones that can separate different faults clearly. Then if needed, these generated features can be combined in some method such as a neural network or clustering at the feature-level. Next, several classifiers are utilized to classify the calculated features or fused features; the diagnostics decisions from all classifiers are grouped as a decision vector and sent into a specific decision-level fusion algorithm to get a more reliable diagnostics decision. 608

5 Figure 5. Flowchart of CBM integrating data fusion Figure 6. Flowchart of fusion diagnostics subsystem. 609

6 B.3. Fusion monitoring and prognostics subsystem There has been much progress in the technology of CBM in recent years. However, many fundamental issues still remain in condition monitoring and prognostics [13]: Indicators, used for accurate condition monitoring and prognostics, need to be developed. Currently, methods are generally focused on solving the failure prediction problems; tools for system performance assessment and degradation prediction have not been well addressed. Most of the developed prognostics approaches are applied on specific equipment. A generic prognostics system is needed. In this subsection, a data fusion based condition monitoring and prognostics subsystem is proposed to solve the above issues. The flowchart of the proposed subsystem is shown in Fig. 7. Condition monitoring is a major component of CBM. The use of condition monitoring allows maintenance to be scheduled or other actions to be taken to avoid the consequences of failure before the failure occurs. Each monitoring indicator has its own merits and shortcomings and is only effective for certain failures at certain stages. Therefore, fusing mutual indicator information is expected to provide accurate information for degradation monitoring. In this proposed subsystem, signals of multi-sensors attached on an operating machine are collected, and features exhibiting a degradation trend are extracted. Then those features are normalized and grouped as an input set for a feature-level fusion algorithm. Next, the process of denoising is considered for filtering process noise from the feature extraction and fusion. After the de-noising process, a clear tracking trend for the operating state can be determined. Finally, condition monitoring can be carried out and a comparison is exerted continuously between the alarm threshold and each indication value. When the monitoring curve crosses through the threshold, a data-driven prognostics module is triggered. Prognostics is the process of predicting the future reliability of a product by assessing the extent of deviation or degradation of the product from its expected normal Figure 7. Flowchart of fusion monitoring and prognostics subsystem. operating conditions [8]. Considering the generality of analysis methods and convenience of monitored system operating data, the data-driven approach is used in this subsystem. Usually, the degradation trend of a machine s performance is reflected in a non-linear or chaotic manner. State space reconstruction is the first step in non-linear chaotic time-series prediction. The reconstruction parameters, delay time, and embedding dimension need to be determined appropriately. Then performance degradation prediction is conducted, which includes not only degradation trend prediction (point estimate) but also the evaluation of uncertainty bounds (interval estimate). Next, the prediction results of several non-linear prediction algorithms can be fused to improve prediction accuracy. Finally, prognostics assessment (PA) is carried out. The core of PA is estimating the remaining useful life of a failing component or system and assigning uncertainty bounds to the degradation trend that will provide maintainers with the earliest and the latest 610

7 (with increasing risk) time to perform maintenance and the associated risk factor when maintenance action is delayed. III. CONCLUSIONS This paper presents an advanced CBM system that integrates RCM for process management and employs data fusion strategy for improvement of maintenance performance. The advantages of the proposed system can be summarized as follows: Cost-effectiveness: the proposed system is constructed under the architecture of RCM, which assists the maintenance manager in identifying vital components, potential failures, proper maintenance tasks, and proper maintenance intervals. It supports an adaptive and dynamic maintenance strategy so that cost-effectiveness can be achieved. Accuracy: use of the data-fusion strategy improves the accuracy of the CBM tool in the process of condition monitoring, diagnostics, and prognostics. Moreover, obtaining accurate maintenance information helps a manager make right maintenance decisions and sustain a continuous and reliable workflow of equipment. Generality: the proposed maintenance system is constructed based on OSA-CBM architecture, which uses RCM management mechanism as maintenance process, and a standard data fusion structure as maintenance improvement. Therefore, general maintenance techniques can be easily embedded into this system. REFERENCES [1] G. Niu, B.S. Yang, Dempster-Shafer regression for multi-step-ahead time-series prediction towards data-driven machinery prognosis, Mechanical System and Signal Processing, Vol. 23 (3), pp , [2] A. Starr, A structured approach to the selection of condition based maintenance, 5th International Conference on FACTORY The Technology Exploitation Process, Cambridge, UK, [3] M. Thurston, M. Lebold, Open standards for condition-based maintenance and prognostic systems, MARCON, [4] E. Zio, Review Reliability Engineering: Old problems and new challenges, Reliability Engineering and System Safety, Vol. 94, pp , [5] L. Matti, On the optimal strategies of condition monitoring and maintenance allocation in distribution systems, 9th International Conference on Probabilistic Methods Applied to Power Systems, Stockholm, Sweden, [6] D.L. Hall, J. Llinas, An introduction to multisensor data fusion, IEEE Digital Object Identifier, Vol. 85 (1), pp. 6-23, [7] J. Moubray, Reliability-Centered Maintenance, 2nd, Butterworth- Heinemann Ltd, London, England, [8] M.G. Pecht, Prognostics and Health Management of Electronics, Wiley-Interscience, New York, NY, [9] G. Carmignani, An integrated structural framework to cost-based FMECA: The priority-cost FMECA, Reliability Engineering & System Safety, Vol. 94 (4), pp , [10] C.M. Apostolakis, The maintenance management framework, Springer London, [11] A.K.S. Jardine, D. Lin, D. Banjevic, A review on machinery diagnostics and prognostics implementing condition-based maintenance, Mechanical Systems and Signal Processing, Vol. 20, pp , [12] O.D. Richard, E.P. Hart, G.S. David, Pattern Classification, Second Edition, Wiley, [13] J. Lee, H. Qiu, J. Ni, D. Djurdjanovi, Infortronics technologies and predictive tools for next- generation maintenance systems, 11th IFAC Symposium on Information Control Problems in Manufacturing,