Condition-based maintenance (CBM) is based on condition monitoring and aims at preforming maintenance based on the system condition and its trend. CBM can be used to realize RCM (Niu &
Pecht, 2009).
Condition monitoring constantly measures and analyses relevant mechanical and electrical component parameters during operation. The parameters selected for monitoring allow determination of the condition and failure state. The need for maintenance of a component is only indicated if parameters show a predefined degradation (Kolerus & Wassermann, 2011).
The difference between CBM and preventive on-condition maintenance is that OC checks a system at defined intervals while condition monitoring continuously monitors the system.
Condition monitoring is used in a wide field of application, including rotary machines (gear boxes, gas and wind turbines, bearings etc. (Mahamad, et al., 2010) (Saravanan & Ramachandran, 2009) (Sugumaran & Ramachandran, 2011) (Tian & Zuo, 2010) (Zhao, et al., 2009), plants and structures (bridges, pipelines etc. (Goode, et al., 2000)). Vibration data are often used to perform the condition monitoring (Ebersbach & Peng, 2008).
The condition of the system is defined by setting limits on certain values based on experience (Mobley, 2002) or on a mathematical or data-driven model (Kolerus & Wassermann, 2011) (Williams, et al., 1994). Machine learning techniques, e.g., decision trees (Sugumaran &
Ramachandran, 2007) (Sugumaran & Ramachandran, 2011) (Tran, et al., 2009), vector support machines (Pham, et al., 2012) (Sugumaran, et al., 2007) (Widodo & Yang, 2007) and neural networks (Chen, et al., 2012) (Mahamad, et al., 2010) (Tian, 2012), are often used to map the features of the input signal to a condition.
Another option is to use a mathematical model, feed the sensor input to the model, calculate the output and check how the output of the theoretical model deviates from the real system. This approach can also be used for fault isolation and identification of failures in addition to prognosis (Wang, et al., 2008) (Williams, et al., 1994) (Kolerus & Wassermann, 2011) (Jardine, et al., 2006).
Data-driven models use past data to create models with stochastic or machine learning algorithms (Pecht, 2008) (Garcia, et al., 2006) (Jardine, et al., 2006). These models require many data samples that represent different conditions of the system. Data-driven models require less human input than mathematical models; model validation and testing can be performed almost automatically.
Trend analysis is a method to achieve CBM. The analysis algorithm looks at recorded parameters at a single moment in time, but takes the full parameter history into account. The need for maintenance of a component is only indicated if the trend of a parameter shows degradation.
Based on the parameter time history, the analysis algorithm can forecast the remaining lifetime of the component (Kolerus & Wassermann, 2011). A variety of methods are suitable for predicting future values. ARMA, ARIMA, artificial neural networks, sequential Monte Carlo and Markov models are used to predict values for a complex time series (Chen, et al., 2011) (Caesarendra, et al., 2010) (Pham & Yang, 2010) (Tian, et al., 2010). The output of the prediction is normally an estimated time to failure (ETTF) and a confidence interval (Sikorska, et al., 2011). The confidence interval defines the reliability of a prediction (Schruben, 1983) (Sikorska, et al., 2011) and can be calculated using a standard time series.
Implementing CBM is both difficult and costly. Many systems have barriers to its use. These barriers include (among others) (Stecki, et al., 2014):
• Inability to predict accurately and reliably the remaining useful life of a machine (prognostics)
• Inability to continually monitor a machine (sensing)
• Inability of maintenance systems to learn and identify impending failures and recommend
what action should be taken (reasoning).
• Initiation of CBM without full knowledge of how the system can fail
• Focusing of CBM research on specific techniques (better mousetrap syndrome)
Condition Monitoring
There are two strategies of monitoring (Randall, 2011) (Kolerus & Wassermann, 2011):
• Permanent monitoring is based on fixed, installed measurement systems. These systems often need to be very complex to react correctly if a failure occurs. They are used if a fast reaction is required after a failure. Permanent monitoring frequently shuts down a machine if a dangerous failure is detected (Randall, 2011).
• Intermittent monitoring is generally used for failure prediction and diagnosis.
Measurements are taken on a regular basis with a mobile device. Data evaluation is done with an external device. Intermittent monitoring is often used to give long-term warnings (Kolerus & Wassermann, 2011).
Permanent monitoring is a better choice than intermittent monitoring when fast reaction times are required, but intermittent monitoring can do more complex computations (Randall, 2011).
Permanent and intermittent monitoring can be combined using the same sensors and working in parallel. This allows intermittent monitoring to be carried out more often (hence, data are always available (Randall, 2011).
Methods of condition monitoring include the following (Randall, 2011):
• Vibration analysis measures the vibration of a machine or system and compares it to a given vibration signature. Vibrations can be linked to events in a machine based on their frequency. Therefore, a vibration signal is often analysed in the time domain and in the frequency domain. Vibration analysis is frequently used for condition monitoring (Randall, 2011) (Kolerus & Wassermann, 2011).
• Lubricant/oil analysis analyses the quality of the fluid and determines whether particles are in it. Contaminants in lubrication oils and hydraulic fluids can lead to the failure of the machine/system. The physical condition of a fluid can be measured in viscosity, water content, acidity and basicity. For a condition monitoring strategy, this means condition-based oil change. It is also possible to detect wear of mechanical systems with a particle analysis (Williams, et al., 1994).
• Performance analysis is an effective way of determining whether a machine is functioning correctly. Performance analysis monitors process parameters such as temperature, pressure, flow rate or processed items per hour (Randall, 2011).
• Thermography is used to detect hot spots in a system or a machine. It is principally used in quasi-static situations.
Condition monitoring can be one-to-one or one-to-many (Williams, et al., 1994). In one-to-one monitoring, a system parameter measured by a sensor is directly forwarded for signal processing and condition monitoring (see Figure 6) independent of the sub-system to which the parameter belongs (Williams, et al., 1994).
Figure 6: One-to-one condition monitoring (Williams, et al., 1994)
In one-to-many monitoring, one sensor is used to give condition information on more than one parameter (see Figure 7). This type of monitoring helps with failure location (Williams, et al., 1994).
Figure 7: One-to-many condition monitoring (Williams, et al., 1994)