1.1 Historical Perspective

With the rapid development of industrial technology, machine tools have become more and more complex in response to the need for higher production quality. While a significant increase in failure rate due to the complexity of machine tools is becoming a major factor which restricts the improvement of production quality and efficiency.

Before 1950, maintenance was basically unplanned, taking place only when breakdowns occurred. Between1950 and 1960, a time-based preventive maintenance (PM) (also called planned maintenance) technique was developed, which sets a periodic interval to perform PM regardless of the health status of a physical asset. In the later 1960s, reliability centered maintenance (RCM) was proposed and developed in the area of aviation. Traditional approaches of reliability estimation are based on the distribution of historical time-to-failure data of a population of identical facilities obtained from in-house tests. Many parametric failure models, such as Poisson, exponential, Weibull, and log-normal distributions have been used to model machine reliability. However, these approaches only provide overall estimates for the entire population of identical facilities, which is of less value to an end user of a facility [1]. In other words, reliability reflects only the statistical quality of a facility, which means it is likely that an individual facility does not necessarily obey the distribution that is determined by a population of tested facilities of the same type. Therefore, it is recommended that on-line monitoring data should also be used to reflect the quality and degradation severity of an individual facility more specifically.

In the past two decades, the maintenance pattern has been developing in the direction of condition-based maintenance (CBM), which recommends maintenance actions based on the information collected through on-line monitoring. CBM attempts to avoid unnecessary maintenance tasks by taking maintenance actions only when there is evidence of abnormal behavior of a physical asset. A CBM program, if properly established and effectively implemented, can significantly reduce maintenance cost by eliminating the number of unnecessary scheduled PM operations.

Prognostics-based maintenance, which is a typical pattern of predictive maintenance (PdM) has been developed rapidly in recent years. Despite the fact that fault diagnosis and prediction are related to the assessment of the status of equipment, and generally considered together, the goals of the decision-making are obviously different. The diagnosis results are commonly used for passive maintenance decision-making, but the prediction results are used for initiative maintenance decision-making. Its goal is minimum use risk and maximum life. By means of fault prediction, the opportune moment from initial defect to functional fault could be estimated. The failure rate of the whole system or some of the components can be modified, so prognostic technology has become a hot research issue. Now fault prediction techniques are classified into three categories according to the recent literature: failure prediction based on an analytical model, failure prediction based on data, and qualitative knowledge-based fault prediction. Artificial-intelligence-based algorithms are currently the most commonly found data-driven technique in prognostics research [1, 2].

Recently, a new generation of maintenance, e-maintenance, is emerging with globalization and fast growth of communication technologies, computer, and information technologies. e-Maintenance is a major pillar in modern industries that supports the success of the integration of e-manufacturing and e-business, by which manufactures and users can benefit from the increased equipment and process reliability with optimal asset performance and seamless integration with suppliers and customers.

1.2 Diagnostic and Prognostic System Requirements

Diagnostics deals with fault detection, isolation, and identification when it occurs. Fault detection is a task to indicate whether something is going wrong in the monitored system; fault isolation is a task to locate the component that is faulty; and fault identification is a task to determine the nature of the fault when it is detected. In recent years, technological development in areas like data mining (DM), data transmission, and databases has provided the technical support for prognostics. Prognostics deals with fault prediction before it occurs. Fault prediction is a task to determine whether a fault is impending and to estimate how soon and how likely it is that a fault will occur. Diagnostics is post-event analysis, and prognostics is prior event analysis. Prognostics is much more efficient than diagnostics in achieving zero-downtime performance. Diagnostics, however, is required when the fault prediction of prognostics fails and a fault occurs.

As a minimum, the basic technical requirements of diagnostics mainly include

1. Sensor location, which has a significant impact on the measurement accuracy.
2. Feature extraction to obtain the parameter which characterizes equipment performance by utilizing signal processing methods including a fast Fourier Transform (FFT) algorithm, a wavelet transform (WT), and so on.
3. Method of fault classification to increase the accuracy of equipment failure classification.

In addition to those technical requirements mentioned above, to specify prognostics accuracy requirements we also need

1. Data on performance degradation, which indicates the decline of equipment performance in the working process.
2. Methods for life prediction to guarantee the safe operation of equipment and improve economic benefits.
3. A confidence interval to estimate the bounds of parameters in the model-based prediction.

Commonly, some aspects of hardware technology, such as the accuracy of sensors, the selection of the location of sensors, and data acquisition provide the technological foundations of prognostics. Also, computer-assisted software techniques, including data transmission, database, and signal processing methods are essential components of a prognostics system.

1.3 Need for Prognostics and Sustainability-Based Maintenance Management

Any organization that owns any large capital assets will eventually face a crucial decision whether to repair or replace those assets, and when. This decision can have far reaching consequences, replacing too early can mean a waste of resources, and replacing too late can mean catastrophic failure. The first is becoming more unacceptable in today's sustainability-oriented society, and the second is unacceptable in the competitive marketplace.

Equipment degradation and unexpected failures impact the three key elements of competitiveness – quality, cost, and productivity [3]. Maintenance has been introduced to reduce downtime and rework and to increase consistency and overall business efficiency. However, traditional maintenance costs constitute a large portion of the operating and overhead expenses in many industries [4]. More efficient maintenance strategies, such as prognostics-based maintenance are being implemented to handle the situation. It is said that prognostics-based maintenance can reduce the maintenance costs by approximately 25% [5]. Generally, machines go through degradation before failure occurs, monitoring the trend of machine degradation and assessing performance allow the degraded behavior or faults to be corrected before they cause failure and machine breakdowns. Therefore, advanced prognostics focuses on performance degradation monitoring and prediction, so that the failures can be predicted and prevented [6].

If large capital assets are analyzed as repairable systems, additional significant information can be incorporated into maintenance optimization models. When these assets break down, but have not yet reached their end-of life, they can be repaired and returned to operating condition. However, sometimes malfunctioning equipment cannot be properly fixed or repaired to its original healthy condition. In this case, the application of prognostics will help solve this problem and avoid irreparable and irreversible damage. Prognostics provides the basic information for a maintenance management system where a maintenance decision is made by predicting the time when the reliability or the remaining life of a facility reaches the maintenance threshold. However, inappropriate maintenance time will result in waste of resources and a heavier environmental load. Nowadays, more efficient maintenance strategies, such as sustainability oriented maintenance management, are put forward. Sustainability-based maintenance management (SBMM) not only benefits manufacturers and customers economically but also improves environmental performance. Therefore, from both environmental and economic perspectives, improving the energy efficiency of maintenance management is instrumental for sustainable manufacturing. SBMM will be one of the important strategies for sustainable development.

1.4 Technical Challenges in Prognosis and Sustainability-Based Maintenance Decision-Making

In order to implement prognostics, three main steps are needed. (i) Feature extraction and selection: feature extraction is the process of transforming the raw input data acquired from mounted or built-in sensors into a concise representation that contains the relevant information on the health condition. Feature selection is the selection of typical features which reflect an overall degradation trend from the extracted features. (ii) Performance assessment: how to effectively evaluate the performance based on the selected features is crucial to prognostics. A good performance assessment method ought to be capable of fusing different information on multiple features for system degradation assessment. (iii) Remaining life prediction: this is a process using prediction models to forecast future performance and obtain the residual useful life of machinery. Remaining life prediction is the most important step in prognostics; it appears to be a hot issue attracting the most attention.

The key point in carrying out intelligent prognostics is the conversion of all kinds of raw data into useful information which indicates the equipment/components performance degradation process. The proposed framework is shown in Figure 1.1, it consists of two modules, a model training module and a real-time prognostics module. The performance assessment model M_E and the life prediction model M_P are the outputs of the model training module, which are employed in the real-time prognostics module. The model training module consists of four major parts: data pre-processing, feature extraction, performance assessment, and remaining life prediction. The real-time prognostics module consists of five components: real-time data acquisition, data pre-processing, feature extraction, performance assessment, and dynamic life prediction. If degradation appears, then early stage diagnosis/prognosis would be conducted.

Several aspects need to be further investigated before prognostics systems can be reliably applied in real-life situations, such as the incorporation of CM data into reliability analyses; the utilization of incomplete trending data; the consideration of effects from maintenance actions and variable operating conditions; the deduction of the non-linear relationship between the measured condition and the actual degradation; the considerations of failure interactions; the accuracy of assumptions and practicability of requirements, as well as the development of performance measurement frameworks. Repair and maintenance decisions for repairable systems are often based on the remaining useful life (RUL), also known as the residual life. Accurate RUL predictions are of interest particularly when the repairable system in question is a large capital asset. In addition, in a business setting, the economic and strategic life for complex and expensive equipment must be taken into account. This can make maintenance decision-making for such systems difficult.

Since environmental issues are involved in maintenance management, the relationship between energy consumption and performance of maintenance facilities should be taken into consideration during the decision-making process. In order to achieve energy reduction in facilities, it is necessary to study the relationship between energy consumption and the performance of maintenance facilities. For example, energy consumption will vary with wear or reliability of maintenance facilities in the use stage.

Existing research on maintenance planning and scheduling to reduce environmental impacts is quite limited. Normally, only one scheduling objective, such as maintenance cost, is solved in the maintenance planning and scheduling problem. Since sustainability is considered in maintenance management, it is necessary to incorporate the energy models of maintenance facilities into the objective function and constraints. Inevitably, energy consumption models of maintenance facilities become more complex and difficult to solve. Optimization methods could be of significant importance to effectively and efficiently solve these “sustainability” challenges. In addition, models and solution approaches are essential to decide on strategic and tactical plans and to ensure that economic, environmental, and societal aspects are balanced. This demands new solution methods and technology to provide the kind of tools that maintenance decision-making needs. For example, improved algorithms should be employed to optimize multiple scheduling objectives, such as maintenance cost and total energy consumption.

The technical challenges of sustainability-based maintenance decision-making mainly consists of three aspects: (i) energy consumption modeling of maintenance facilities, (ii) establishing the relationship between energy consumption and performance of maintenance facilities, and (iii) solving the sustainability-based maintenance planning and scheduling problem.

In order to propose efficient and realistic strategies for reducing the consumption of energy and resources, it is imperative to develop methods for estimating the energy consumption of maintenance facilities. Maintenance can manage product quality and quality of services during the use phase. It also decreases environmental impacts since equipment in good condition can use energy efficiently and its physical life can be extended. However, when a maintenance system is not properly constructed, the efficiency of maintenance can be lower and might harm life-cycle management. Moreover, energy consumption models are the inputs of sustainability-based maintenance planning and scheduling problems. It is, therefore, important to establish reliable energy consumption models of maintenance facilities with a high accuracy.

1.5 Data Processing, Prognostics, and Decision-Making

Data acquisition, data processing, prognostics, and maintenance decision-making are the four key elements of a prognostics-based maintenance management flowchart (see Figure 1.2).

Data acquisition is the process of collecting, converting, and recording useful data from targeted physical assets. The hardware of data acquisition systems typically includes sensors, an amplifier circuit, an analog-to-digital (A/D) converter, a data transmission device, and a data recording circuit. A sensor is a converter that measures a physical quantity and converts it into a signal which can be read by an observer or by a (nowadays mostly electronic) instrument. An electronic amplifier, amplifier, or (informally) amp is an electronic device that increases the power of a signal by taking energy from a power supply and controlling the output to match the input signal shape but with a larger amplitude. An A/D converter is a device that converts a continuous physical quantity (usually voltage) to a digital number that represents the quantity's amplitude. In real-time monitoring systems, the control computers are far from the targeted assets. The digital signals indicating the health state of the assets need to transmit from the on-site plant to the control computer.

Data processing plays a crucial role in machinery prognostics and maintenance management. The first step of data processing is data cleaning. This is an important step since data, especially event data, which is usually entered manually, always contains errors. Data cleaning ensures, or at least increases the chance, that clean (error-free) data are used for further analysis and modeling. Without the data cleaning step, one may get into the so-called “garbage in garbage out” situation. Data errors are caused by many factors including the human factor mentioned above. For condition monitoring data, data errors may be caused by sensor faults. In this case, sensor fault isolation is the right way to go. In general, however, there is no simple way to clean data. Sometimes it requires manual examination. Graphical tools would be very helpful in finding and removing data errors. The next step of data processing is data analysis. A variety of models, algorithms, and tools are available in the literature to analyze data for better understanding and interpretation. The models, algorithms, and tools used for data analysis depend mainly on the types of data collected.

Data processing for waveform and multidimensional data is also called signal processing. Various signal processing techniques have been developed to analyze and interpret waveform and multidimensional data to extract useful information for further diagnostic and prognostic purposes. The procedure of extracting useful information from raw signals is the so-called feature extraction.

There are numerous signal processing techniques and algorithms in the literature for diagnostics and prognostics of mechanical systems. Case-dependent knowledge and investigation are required to select appropriate signal processing tools from among a number of possibilities.

The most common waveform data in condition monitoring are vibration signals and acoustic emissions. Other waveform data are ultrasonic signals, motor current, partial discharge, and so on. In the literature, there are three main categories of waveform data analysis: time-domain analysis, frequency-domain analysis, and time–frequency analysis.

The real-time monitoring systems provide fundamental information representing the health states of the monitored systems. The information helps to identify if the asset health has deviated from the normal. Then fault diagnostics and prognostics can be implemented. Fault diagnostics is used to detect, isolate, and identify the abnormal phenomenon. However, the more important question is how to utilize the health information to predict how long the machine can operate safely and perform its function, in order to optimize the maintenance schedules and ultimately maximize organizational efficiency. That is the relatively new research topic – prognostics which provides critical information such as early stage fault recognition and remaining life prediction for diagnostics.

Prognostics, the real issues involved with predicting life remaining, have been defined in the literature. ISO 13381–1(3) [7] defines prognosis as a “Technical process resulting in determination of remaining useful life”. Jardine et al. [8] define two main prediction types in machine prognosis. The most widely used prognosis is “To predict how much time is left before a failure (or, one or more faults) occurs given the current machine condition and past operation profile”. The time left before observing a failure is usually called remaining useful life or sometimes just the term useful life is used. The second prediction type is for situations when a failure is catastrophic (e.g., in nuclear power plants). The probability that a machine operates without a failure up to the next inspection interval, when the current machine condition and the past operation profile are known, is predicted. Damage prognosis is a frequently used term in structural safety and reliability. It is defined, as the estimate of an engineered system's remaining useful life [9].

Rule-based prognostic systems detect and identify incipient faults in accordance with the rules representing the relation of each possible fault to the actual monitored equipment condition. Case-based prognostic systems use historical records of maintenance cases to provide an interpretation for the actual monitored conditions of the equipment. The case library of maintenance is required to record all previous incidents, faults, and malfunctions of equipment which are used to identify the historical case that is most similar to the current condition. If a previous equipment fault occurs again, a case-based prognostic system will automatically pick up the maintenance advice, including trouble–cause–remedy, from the case library. A model-based prognostic system uses different mathematical, neural network, and logical methods to improve prognostic reasoning based on the structure and properties of the equipment system. A model-based prognostic system compares the real monitored condition with the model of the object in order to predict the fault behavior.

Maintenance managers deal with manufacturing systems that are subject to deteriorations and failures. One of their major concerns is the complex decision-making problem when they consider the availability aspect as well as the economic issue of their maintenance activities. They are continuously looking for a way to improve the availability of their production machines in order to ensure given production throughputs at the lowest cost.

This decision-making problem concerns the allocation of the right budget to the appropriate equipment or component. The objective is to minimize the total expenditure and to maximize the effective availability of production resources.

Proper instrumentation of critical systems and equipment plays a vital role in the acquisition of necessary technical data, while the support of analytical software with embedded mathematical models is crucial for the decision-making process.

The intelligent predictive decision support system (IPDSS) for maintenance integrates the concepts of:

1. Equipment condition monitoring.
2. Intelligent condition-based fault diagnosis.
3. Prediction of the trend of equipment deterioration.

Through integrating these three elements, the quality of maintenance decisions could be improved.

1.6 Sustainability-Based Maintenance Management

SBMM is a maintenance program that implements maintenance actions (diagnostics, PM, CBM, and prognostics) to obtain sustainability oriented maintenance strategies that minimize negative environmental impacts, conserve energy and natural resources, are safe for employees, communities, and consumers and increase the availability, reliability, and life span of facilities to keep high productivity and reduce maintenance cost, which will make maintenance actions balance with respect to economic, environmental, and societal aspects.

A traditional maintenance management system includes decision processes, such as selection of end of life options, including reuse and recycle. Decisions are made based on the conditions of the products which are subject to maintenance. If such SBMM works, component reuse can be promoted.

Figure 1.3 shows the relation between maintenance and manufacturing. Maintenance can improve the quality of the collected components and reduce the work required for quality assurance.

A SBMM system can be interpreted as a life-cycle management process. It includes environment-based maintenance service providers and a monitoring system connected to the equipment. A SBMM system, if properly established and effectively implemented, can significantly reduce maintenance cost, environmental burden, and societal impacts to improve the competitiveness of an enterprise.

The concept of SBMM has become increasingly important as a measure to reduce environmental impact and resource consumption in manufacturing. Figure 1.4 shows the circular manufacturing with maintenance in the product use stage. As depicted, the life-cycle options, such as maintenance, upgrade, reuse, and recycling, which correspond to various paths in circular manufacturing, are means to reduce environmental load and resource consumption.

We use technologies such as condition diagnosis, residual life estimation, disassembly, restoration (including cleaning, adjustment, repair, and replacement), inspection, and re-assembly to achieve maintenance management. When products continue to be used by the same user, the activities to maintain or enhance the original functionality of the product are called maintenance and upgrade. The maintenance technologies are necessary to exhaust the item's life to the fullest extent possible through restoration and upgrade.

The maintenance strategies in Figure 1.4 have been selected without regard for reuse, and reuse has been discussed without regard for recycling. To make effective use of maintenance, we need to plan the product life-cycle maintenance management [10]. For example, reused products should be recycled at the end. According to the concept of the product life-cycle planning, the implementation of life-cycle options should be discussed in an integrated way. However, in the conventional architecture of maintenance management illustrated in Figure 1.4, maintenance, reuse, and recycling are represented as supplemental processes.

On the basis of the recognition that the purpose of life-cycle maintenance is to provide the required function to users, there is no reason to discriminate between newly produced products and reused products as far as they satisfy user needs. In this sense, we should integrate maintenance into life-cycle manufacturing as indicated in Figure 1.5. We call such a system life-cycle maintenance management because the innermost loop that is, maintenance, is prioritized as the most efficient circulation.

1.7 Future of Prognostics-Based Maintenance

The definition of prognostics has already been put forward and prognostic techniques are developing rapidly in some areas. However, prognostics-based maintenance still needs further research, in particular:

1. The development of smart sensors and other low-cost on-line monitoring systems that will permit the cost-effective continuous monitoring of key equipment items. An example is the micro-electro-mechanical sensor (MEMS), an accelerometer that is produced in silicon using the same processes as integrated circuit manufacture. It allows the sensor and amplifier electronics to be integrated into a single chip to replace traditional piezoelectric accelerometers.
2. The increasing provision of built-in sensors as standard features in large motors, pumps, turbines, and other large equipment and critical components.
3. The development of fusion techniques in the complete maintenance to improve overall reliability.
4. Increasing integration and acceptance of common standards for integrating maintenance software. A general platform needs these standards to share information, transfer data, make decisions, and so on.

Diagnostic and preventive maintenance are not, however, the terminal goals of our research and obviously will not meet the fast development of high-tech in the near future. For the sake of higher flexibility and lower maintenance cost, biotechnology is the main area to consider for future scientific research. Bio-mechanisms of self-recovery and self-healing are worth further research and will have broad application prospects in maintaining the performance of equipment. By utilizing biotechnology, prognostics-based maintenance will eventually implement true continuous production.

References

1. 1. Heng, A., Zhang, S., Tan, A.C.C. et al. (2009) Rotating machinery prognostics: state of the art, challenges and opportunities. Mechanical Systems and Signal Processing, 23, 724–739.
2. 2. Lee, J., Ni, J., Djurdjanovic, D. et al. (2006) Intelligent prognostics tools and e-maintenance. Computers in Industry, 57, 476–489.
3. 3. Yan, J., Koc, M. and Lee, J. (2004) A prognostic algorithm for machine performance assessment and its application. Production Planning and Control, 15, 796–801.
4. 4. Gebraeel, N. and Lawley, M. (2004) Residual life predictions from vibration-based degradation signals: a neural network approach. IEEE Transactions on Industrial Electronics, 51, 694–699.
5. 5. Camci, F. (2005) Process monitoring, diagnostics and prognostics using support vector machines and Hidden Markov Model. PhD thesis. Department of Industry Engineering, Wayne State University, Detroit, Michigan.
6. 6. Huang, R., Xi, L., Li, X. et al. (2007) Residual life predictions for ball bearings based on self-organizing map and back propagation neural network methods. Mechanical Systems and Signal Processing, 21, 193–207.
7. 7. ISO ISO 13381–1. (2004) Condition Monitoring and Diagnostics of Machines Prognostics Part1: General Guidelines, International Organization for Standardization, Geneva.
8. 8. Jardine, A.K.S., Lin, D. and Banjevic, D. (2006) A review on machinery diagnostics and prognostics implementing condition-based maintenance. Mechanical Systems and Signal Processing, 20, 1483–1510.
9. 9. Farrar, C.R. and Lieven, N. (2006) Damage prognosis: the future of structural health monitoring. Philosophical Transactions of the Royal Society, Series A, 365, 623–632.
10. 10. Umeda, Y., Takata, S., Kimura, F. et al. (2012) Toward integrated product and process life cycle planning—An environmental perspective. CIRP Annals—Manufacturing Technology, 61 (2), 681–702.