3.2.1Maintenance Paradigm Overview
Looking back on the development history and forecasting the development tendency of maintenance technologies, the roadmap to excellence in maintenance can be illustrated as in Figure 3.1.
3.2.1.1No Maintenance
There are two kinds of situations in which no maintenance will occur:
• No way to fix it: the maintenance technique is not available for a special application, or the maintenance technique is at too early stage of develop- ment.
• Isn’t worth it to fix it: some machines were designed to be used only once.
When compared to maintenance cost, it may be more cost-effective just to discard it.
Neither of the scenarios above is within the scope of the discussion here.
Machine Performance and uptime
No Maintenance
Preventive Maintenance
(Scheduled Maintenance) Reactive
Maintenance (Fire Fighting)
Predictive Maintenance
Proactive Maintenance (Failure Root causes analysis)
Self-Maintenance or Maintenance-free
Machine
Figure 3.1. The development of maintenance technologies
3.2.1.2Reactive Maintenance
The aim of reactive maintenance is just to “fix it after it’s broken”, since most of the time a machine breaks down without warning and it is urgent for the maintenance crew to put it back to work: this is also referred to as “fire-fighting”.
This fire-fighting mode of maintenance is still present in many maintenance operations today because accurate knowledge of the equipment behavior is lacking.
Essentially, little to no maintenance is conducted and the machinery operates until a failure occurs. At this time, appropriate personnel are contacted to assess the situation and make the repairs as expeditiously as possible. In a situation where the damage to equipment is not a critical factor, plenty of downtime is available, and the values of the assets are not a concern, the fire-fighting mode may prove to be an acceptable option. Of course, one must consider the additional cost of making repairs on an emergency basis since soliciting bids to obtain reasonable costs may not be applicable in these situations. Due to market competition and environmental/safety issues, the trend is toward appropriating an organized and efficient maintenance program as opposed to firefighting.
3.2.1.3Preventive Maintenance
Preventive maintenance (PM) is an equipment maintenance strategy based on replacing, overhauling or remanufacturing an item at fixed or adaptive intervals, regardless of its condition at the time. These maintenance operations models can be characterized as long term maintenance policies (Wang 2002) that do not take into account instantaneous equipment status. Scheduled restoration tasks and scheduled discard tasks are both examples of preventive maintenance tasks.
In preventive maintenance, breakdowns are tracked and recorded in a database, and the information accumulated provides a base for general preventive actions.
The age-dependent PM policy can be considered as the most common maintenance policy in which a unit’s PM times are based on the age of the unit. The basic idea is to replace or repair a unit at its age T or failure whichever occurs first (Badia et al., 2002; Mijailovic 2003). Commonly used equipment reliability indices such as mean time between failure (MTBF) and mean time to repair (MTTR) are extracted
from the historical databases of equipment behavior over time. These two indices provide a rough estimate of the time between two adjacent breakdowns and the mean time needed to restore a system when such breakdowns happen. Although equipment degradation processes vary from case to case, and the causes of failure can be different as well, the information contained in MTBF and MTTR can still be informative. Other indices can also be extracted and used, including the mean lifetime, mean time to first failure, and mean operational life, as discussed by Pham et al. (1997). With the introduction of minimal repair and imperfect maintenance, various extensions and modifications to the age-dependent PM policy have been proposed (Bruns 2002; Chen et al. 2003). Another preventive maintenance policy that received much attention is the periodic PM policy, in which degraded machines are repaired or replaced at fixed time intervals independent of the equipment failures. Various modifications and enhancements to this maintenance policy have also been proposed recently (Cavory et al. 2001).
The preventive maintenance schemes are time-based without considering the current health state of the product, and thus are inefficient and less valuable for a customer whose individual asset is of the most concern. For the case of helicopter gearboxes, it was found that almost half of the units were removed for overhaul even though they were in a satisfactory operating condition. Therefore techniques for more economical and reliable maintenance are needed.
3.2.1.4Predictive Maintenance
Predictive maintenance (PdM) is a right-on-time maintenance strategy. It is based on the failure limit policy in which maintenance is performed only when the failure rate, or other reliability indices, of a unit reaches a predetermined level. This maintenance strategy has been implemented as condition based maintenance (CBM) in most production systems, where certain performance indices are periodically (Barbera et al. 1996; Chen and Trivedi 2002) or continuously monitored (Marseguerra et al.
2002). Whenever an index value crosses some predefined threshold, maintenance actions are performed to restore the machine to its original state, or to a state where the changed value is at a satisfactory level in comparison to the threshold.
Predictive maintenance can be best described as a process that requires both technology and human skills, while using a combination of all available diagnostic and performance data, maintenance history, operator logs and design data to make timely decisions about maintenance requirements of major/critical equipment. It is this integration of various data, information and processes that leads to the success of a PdM program. It analyzes the trend of measured physical parameters against known engineering limits for the purpose of detecting, analyzing and correcting a problem before a failure occurs. A maintenance plan is devised based on the prediction results derived from condition based monitoring. This method can cost more up front than PM because of the additional monitoring hardware and software investment, cost of manning, tooling, and education that is required to establish a PdM program. However, it provides a basis for failure diagnostics and maintenance operations, and offers increased equipment reliability and a sufficient advance in information to improve planning, thereby reducing unexpected downtime and operating costs.
3.2.1.5Proactive Maintenance
Proactive maintenance (PaM) is a new maintenance concept that is emerging along with the development of business globalization. It encompasses any tasks that seek to realize the seamless integration of diagnosis and prognosis information and maintenance decision making via a wireless internet or satellite communication network. Machine health information should represent a trend, not just a status, so that a company’s productivity can be focused on asset-level utilization, not just production rates. Moreover, through integrated life-cycle management, such degradation information can be used to make improvements in every aspect of a product’s life-cycle. Intelligent maintenance systems (IMS) presented by Lee (1996) is a PaM representative. Specifically, it has three main working directions as follows:
• Develop intertwined embedded informatics and electronic intelligence in a networked and tether-free environment and enable products and systems to intelligently monitor, predict, and optimize their performance.
• Change “failure reactive” to “failure proactive” by avoiding the underlying conditions that lead to machine faults and degradation. Focus on analyzing the root cause, not just the symptoms. That is, seek to prevent or to fix failure from its source.
• Feed the maintenance information back to the product, process and machine design, and ultimately make improvements in every aspect of product life- cycle.
3.2.1.6Self-maintenance
Self-maintenance is a new design and system methodology. Self-maintenance machines are expected to be able to monitor, diagnose, and repair themselves in order to increase their uptime.
One system approach to enabling self-maintenance is based on the concept of functional maintenance (Umeda et al. 1995). Functional maintenance aims to recover the required function of a degrading machine by trading off functions, whereas traditional repair (physical maintenance) aims to recover the initial physical state by replacing faulty components, cleaning, etc. The way to fulfil the self-maintenance function is by adding intelligence to the machine, making it clever enough for functional maintenance, so that the machine can monitor and diagnose itself, and it can still maintain its functionality for a while if any kind of failure or degradation occurs. In other words, self-maintainability would be appended to an existing machine as an additional embedded reasoning system. The required capabilities of a self-maintenance machine (SMM) are defined as follows (Labib 2006):
• Monitoring capability: SMM must have the ability of on-line condition monitoring using sensor fusion. The sensors send the raw data of machine condition to a processing unit.
• Fault judging capability: from the sensory data, the SMM can judge whether the machine condition is at normal or abnormal state. By judging the condition of the machines, we can know the current condition and time left to failure of the machines.
• Diagnosing capability: if the machine condition is at abnormal state, the causes of faults must be diagnosed and identified to allow repair planning action to be carried out.
• Repair planning capability: the machine is able to propose repair actions based on the result of diagnosis and functional maintenance. The repair planning action is performed using knowledge from the experts which is stored in the data base system. There may be more than one repair action proposed; however, the optimized one will be selected to be implemented.
• Repair executing capability: the maintenance is carried out by the machine itself without any human intervention. This can be achieved through computer control system and actuators in the machines.
• Self-learning and improvement: when faced with unfamiliar problems, the machine is able to repair itself and it is expected that if such problems occur again, the machine will take a shorter time for repairing itself and the outcome of maintenance will be more effective and efficient.
Efforts towards realizing self-maintenance have been mainly in the form of intelligent adaptive control, where investigation of control was achieved using fuzzy logic control. In order to realize self-maintenance, one needs to develop and implement an adaptive artificial neuron-fuzzy inference system which allows the fuzzy logic controller to learn from the data it is modeling and automatically produce appropriate membership functions and the required rules. Such a controller must be able to cater for sensor degradation and this leads to self-learning and im- provement capabilities.
Another system approach to enabling self-maintenance is to add the self-service trigger function to a machine. The machine self-monitors, self-prognoses and self- triggers a service request before a failure actually occurs. The maintenance task may still be conducted by a maintenance crew, but the no gap integration of machine, maintenance schedule, dispatch system and inventory management system will minimize maintenance costs and raise customer satisfaction.
3.2.2Prognostics Approaches for Condition Based Maintenance
Condition based maintenance (CBM) was presented as a maintenance scheme to provide sufficient warning of an impending failure on a particular piece of equip- ment, allowing that equipment is to be maintained only when there is objective evidence of an impending failure. CBM methods and practices have been con- tinuously improved in recent decades. Sensor fusion techniques are now commonly in use due to the inherent superiority in taking advantage of mutual information from multiple sensors (Hansen et al. 1994; Reichard et al. 2000; Roemer et al.
2001). A variety of techniques in vibration, temperature, acoustic emissions, ultrasonic, oil debris, lubricant condition, chip detectors, and time/stress analyses has received considerable attention. For example, vibration signature analysis, oil analysis and acoustic emissions, because of their excellent capability for describing machine performance, have been successfully employed for prognostics for a long time (Kemerait 1987; Wilson et al. 1999; Goodenow et al. 2002). Current prognostic approaches can be classified into three basic groups: model-based
approach, data-driven approach, and hybrid approach. The model-based approach requires detailed knowledge of the physical relationships between, and characteris- tics of, all related components in a system. It is a quantitative model used to identify and evaluate the difference between the actual operating state determined from measurements, and the expected operating state derived from the values of the characteristics obtained from the physical model. Bunday (1991) presented the theory and methodology of obtaining reliability indices from historical data. In direct implementation in maintenance, the reliability of the system is kept at a defined level, and whenever the reliability falls below the defined level, main- tenance actions should take place to restore it back to its proper level. However, it is usually prohibitive to use the model-based approach since relationships and characteristics of all related components in a system and its environment are often too complicated to build a model with a reasonable amount of accuracy. In some cases, values of some process parameters/factors are not readily available. A poor model leads to poor judgment. The data-driven approach requires a large amount of history data representing both normal and “faulty” operations. It uses no a priori knowledge of the process but, instead, derives behavioral models only from measurement data from the process itself. Pattern recognition techniques are widely used in this approach. General knowledge of the process can be used to interpret data analysis results, based on which qualitative methods such as fuzzy logic, and artificial intelligence methods can be used for decision making to realize fault prevention. The hybrid approach fuses the model-based information and sensor-based information and takes advantage of both model-driven and data- driven approaches through which more reliable and accurate prognostic results can be generated (Hansen et al. 1994). Garga et al. (2001) introduced a hybrid reasoning method for prognostics, which integrated explicit domain knowledge and machinery data. In this approach, a feed-forward neural network was trained using explicit domain knowledge to get a parsimonious representation of the explicit domain knowledge.
However, a major breakthrough has not been made since. Existing prognostic methods are application or equipment specific. For instance, the development of neural networks has added new dimensions to solving existing problems in con- ducting prognostics of a centrifugal pump case (Liang et al. 1988). A comparison of the results using the signal identification technique shows various merits of employing neural nets including the ability to handle multivariate wear parameters in a much shorter time. A polynomial neural network was conducted in fault detec- tion, isolation, and estimation for a helicopter transmission prognostic application (Parker et al. 1993). Ray and Tangirala (1996) built a stochastic model of fatigue crack dynamics in mechanical structures to predict remaining service time. Fuzzy logic-based neural networks have been used to predict paper web breakage in a paper mill (Bonissone 1995) and the failure of a tensioned steel band with seeded crack growth (Swanson 2001). Yet another prognostic application presented an integrated system in which a dynamically linked ellipsoidal basis function neural network was coupled with an automated rule extractor to develop a tree-structured rule set which closely approximates the classification of the neural network (Brotherton et al. 2000). That method allowed assessment of trending from the nominal class to each of the identified fault classes, which means quantitative
prognostics were built into the network functionality. Vachtsevanos and Wang (2001) gave an overview of different CBM algorithms and suggested a method to compare their performance for a specific application.
Prognostic information, obtained through intelligence embedded into the manufacturing process or equipment, can also be used to improve manufacturing and maintenance operations in order to increase process reliability and improve product quality. For instance, the ability to increase reliability of manufacturing facilities using the awareness of the deterioration levels of manufacturing equipment has been demonstrated through an example of improving robot reliability (Yamada and Takata 2002). Moreover, a life cycle unit (LCU) (Seliger et al. 2002) was proposed to collect usage information about key product components, enabling one to assess product reusability and facilitating the reuse of products that have significant remaining useful life.
In spite of the progresses in CBM, many fundamental issues still remain. For example:
1. Most research is conducted at the single equipment level, and no infra- structure exists for employing a real-time remote machinery diagnosis and prognosis system for maintenance.
2. Most of the developed prognostics approaches are application or equipment specific. A generic and scalable prognostic methodology or toolbox doesn’t exist.
3. Currently, methods are focused on solving the failure prediction problem.
The need for tools for system performance assessment and degradation prediction has not been well addressed.
4. The maintenance world of tomorrow is an information world for feature- based monitoring. Features used for prognostics need to be further de- veloped.
5. Many developed prediction algorithms have been demonstrated in a labo- ratory environment, but are still without industry validation.
To address the afore-mentioned unmet needs, Watchdog Agent®-based intelligent maintenance systems (IMS) has been presented by the IMS Center with a vision to develop a systematic approach in advanced prognostics to enable products and systems to achieve near-zero breakdown reliability and performance.