A Methodology for Selection of Condition Monitoring Techniques for Rotating Machinery

Rotating machinery generally consist of a driver machine such as a motor and a driven machine or load such as a compressor or pump. Several condition monitoring (CM) techniques have been developed over the years for the predictive maintenance of rotating machinery. An appropriate selection of these techniques needs to be established for maximizing the ROI (Return on investment) of such systems. This paper proposes a methodology for the proper selection of CM techniques based on factors such as fault detectability, fault severity, cost, ease of data collection, noise, and system criticality. Effective techniques are recommended based on applicability in the industrial scenario and research done till now. A careful scoring system was adopted and weightage was given to each factor by expert opinion depending on its importance in the industrial environment. Multi-criteria decision-making (MCDM) was used to obtain comparable technique combination scores. The effectiveness of a single technique was found limited in rotating machinery, effective combinations were made and scored according to important factors. Final scores were obtained and top combinations were chosen for non-critical, sub-critical, and critical systems. A possible way of implementation is also shown for remote monitoring through literature.


INTRODUCTION
A motor connected to a load is the most common form of industrial asset. Especially Induction motors (IM) consume nearly 30-40% of the total world's energy, and 20% of total energy is used in systems for moving fluids in which pumps especially centrifugal pumps are mostly used in many industries as a load (Resa et al., 2019), (Stopa et al., 2014). The majority of the AC motors are IM and 90% of motors used in industries are IM. The 3-phase IM is more common in the industry, being more efficient than a single-phase motor (Kuphaldt & Haughery, 2000).
IM is so commonly used in industries, as it has so many advantages such as less frequent maintenance requirements, IM has a few things going towards them like robustness, low cost, low maintenance, load handling, and speed control (A. .
Pumping systems are utilized in a variety of industries and are capable of performing a wide range of tasks, making them highly sought after. Industrial pumps, which are one of the different types of pumping systems, are in high demand in industries including oil and gas, power, and food and beverage.
This paper addresses the real-time application and common faults for the most commonly used rotating machinery having IM and a common load such as pumps, compressors, fans, and conveyors. Many factors which are like hurdles to implementing the solution were considered. Researchers have focused on fault detection by suggesting methods and tools for IM and pumps separately (Djeddi et al., 2007;Kanovic et al., 2013;Gugaliya et al., 2018;Mehala & Dahiya, 2007;Ye & Wu, 2000;Jin et al., 2016;Glowacz & Glowacz, 2017;Glowacz, 2018;Vitek et al., 2011;Goktas et al., 2017;Dutta et al., 2018;Stopa et al., 2014;Henriquez et al., 2014;Goman et al., 2019), but this paper will suggest an algorithm to determine the latest technique and sensors to cover both motor and load side. Also suggesting how to implement it with the latest technologies like the Internet of Things (IoT), wireless monitoring, online CM, networking multiple setups for remote monitoring.
The estimates say that 23.9% of total manufacturing cost goes towards downtime cost and 13.3% goes towards planned production time with added hourly downtime cost, planned or unplanned is high enough to be unseen (Tabikh, n.d.).
As per statistics of many industrial settings, the unplanned downtime is much higher than the cost of scheduled downtime, typically the average hourly cost of unplanned downtime is 40K USD. The total estimated cost for industrial manufacturers tops 50 billion USD per year. Some examples of downtime costs for industries are shown in Table 1 (https://Behrtech.Com/Blog/Infographic-20-Mind Boggling-Stats-on-Cost-of-Industrial-Downtime/, n.d.). Fortune 1000, 82% of companies having unplanned downtime in 3 years had an average of 4-hour downtime per failure, costing an average of 2 million USD. Unplanned downtime does not only cost money but also customer trust and productivity (Elliot, 2015).

Downtime costs incurred across some industries Automobile
22000 USD lost every minute of downtime Mining 5 million USD for losing an excavator for a day Oil & Gas 38 million USD of financial loss due to unplanned downtime annually Process Industries 5% of total output value loss due to unplanned downtime Table 1. Downtime cost across some industries (https://Behrtech.Com/Blog/Infographic-20-Mind-Boggling-Stats-on-Cost-of-Industrial-Downtime/, n.d.) Usually, the failure of IM is not sudden but gradual degradation or faulty parts. The efficiency of the motor keeps decreasing due to the occurrence of faults and if remains unchecked it eventually fails. Fault can occur from many reasons such as natural wear, incorrect installation, broken parts, overheating, stress, and much more. Looking at the part failure percentage of IM it is clear that which parts fail the most and how to monitor the motor precisely by planning and scheduling maintenance becomes important (Parekh, 2003). Figure 1 shows the part failure probability data and common loads connected to IM. Common faults from the most used loads with a motor will be the priority in this paper as shown in Figure 2. The other loads connected with the IM have faults that are either non-diagnosable with CM techniques or non-significant to be a concern to the industry. Cavitation in pumps will be considered in our methods with other common faults on the load side (Terron-Santiago et al., 2021;Goman et al., 2019).
We have to effectively plan the maintenance for when there is a change in the efficiency of IM. To catch the fault at the incipient level, CM tools and techniques help to ensure the detection of the fault. We can schedule the maintenance or order spare parts beforehand to avoid any downtime costs.  There are additional operational costs (e.g., electricity bill) due to reduced efficiency based on the Duty Cycle, common electric motor duty cycles are given in There is a need for a common methodology that incorporates cost as well as other factors while selecting CM techniques. (Buckley, 1987;Carnero, 2009;Liu et al., 2016;Mechefske & Wang, 2001;Moore & Baker, 1969;Parsaei & Wilhelm, 1989;Petkov et al., 2020;Utne et al., 2012;Wang & Wang, 2013) Common Motor Duty Cycle as per Importer Exporter Code (IEC) Standards S1 Continuous Running Constant load operation of sufficient duration to reach thermal equilibrium S2 Short-time duty Constant load for a specific time but less than that to reach thermal equilibrium followed by rest to reach coolant temp.

S3 Intermittent
Periodic duty Identical duty cycles sequentially with constant load and rest without a connection This paper proposes a methodology for more effective implementation of CM techniques for industrial applications. A complete solution considering motor as well as load side fault situations is suggested. The article will cover Fault Diagnosis as well as Fault severity monitoring with the proper selection of sensors or their combination. New possibilities of how the techniques can be used will also be discussed in this paper so to help the engineers collect data easily and safely. Figure 3 shows the complete asset schematic system. There are two types of modes in which we apply CM, Online-Applied when the machine is in working condition (e.g., vibration analysis, current analysis, thermography), Offline-Applied when the machine is not working (e.g., checking misalignment) Data collection is done on a continuous and periodic basis, generally, the critical machines need to be monitored continuously due to the high-cost risk and safety hazards involved. For general-purpose machine's periodic data collection method will be well suited where preventing a failure will lead to profit on investment (Laws & Muszynska, 1987). Machine-mounted sensors with the integrated system will give real-time data in case of continuous monitoring and periodic analyses of signals collected in data loggers are essential.
The rest of the paper is organized as follows: Section 2 explains common CM techniques like vibration, current, thermal, acoustic, and flux. Flux sensors are relatively new in the field of CM, with very less research done in this area. Different types of faults that can be detected by these techniques and their effectiveness in fault detection are also shown. Section 3 proposes a methodology for proper technique selection, considering important factors essential for real-time industrial applications, and also proposing a wireless setup for multi-sensor industrial asset applications. Section 4 concluded the results and findings of the paper.
CM techniques will be discussed in Section 2 with their capabilities to detect different types of faults.

CM TECHNIQUES
In this section various CM techniques, their effectiveness in detecting different types of faults, and severity levels of diagnosis have been presented.

Vibration Monitoring
Vibration analysis works on the directional measurement of vibration signals which are collected by data acquisition systems through sensors (accelerometer). This technique is used to detect faults like misalignment, imbalance, bearing failure, cavitation, gear faults, and eccentricity (Han & Song, 2003). The possibility of detecting stator winding faults, uneven air gaps, unbalances in drive load, and asymmetrical power supply when the sensor is placed on the stator is an advantage (Thorsen & Dalva, 1998). Vibration in any machine is not desirable. It can be used to detect faults in the early stages, careful understanding and correct application are essential for a maintenance engineer (Soother & Daudpoto, 2019). Figure 4 shows the vibration severity levels for determining a machine's health ISO:10816-6 1995. It is widely used to determine if the machine requires maintenance.
Vibration measurements can be done in either radial or axial directions (Yu, 2020). If there is a change in the flux distribution of the motor it will cause a change in the spectrum of vibration, this change can be measured to get results regarding the type of fault and severity level of fault. The faulty signal can be compared to a reference point (healthy spectrum) (Gundewar & Kane, 2021). Changing the placement of the sensor can be used to identify different type of faults.
Vibration signal analysis is very useful tool especially in the case of mechanical systems in rotating machines. Unplanned downtime, maintenance costs can be reduced significantly with proper CM using vibration signals. Bearing faults can happen by a defective race, cage, or ball with single or multiple fault locations. A vibration sensor is attached to the bearing to collect vibration data that is sent to the data acquisition system and analyzed in the signal processing software. The healthy or newly commissioned machine vibration spectrum is compared with the faulty spectrum to find changes in the machine's health. Crossreferencing the difference in vibration spectrum with the characteristic fault frequency, we can get an idea of what part of the bearing is defective. Depending on the amplitude of fault frequency, the severity of the fault can be deduced by the International Organization for Standardization (ISO) severity chart Figure 5. Vibrations are produced at every rotation of the rolling elements. Localized faults on every impact cause a series of vibrations, the position, and amplitude of vibrations for every speed can be calculated by knowing bearing dimensions and rotational speed. These are called characteristic fault frequencies (CFC) and are different for every part of the bearing. By cross-referencing each CFC with impulses generated by faulty bearings we can detect which part of the bearing is faulty using mechanical vibration analysis techniques (Djeddi et al., 2007).
Cage Characteristic Fault frequency, Outer Race Characteristic Fault frequency, Inner Race Characteristic Fault frequency, Ball Characteristic Fault frequency, Where is rotational frequency, is ball diameter, is ball contact angle, is ball pitch diameter, is the number of balls.
The CFC that should be observed for detecting bearing faults can be calculated by equations (1-4) (Gugaliya et al., 2018) also some typical bearing faults can be seen in Figure  5.
Rotor bar Faults (breakage) are the main fault of the rotor in the IM as shown in Figure 6. Breakage of one bar increases the stress on other nearby bars which deteriorates their health as well. Generally, vibration monitoring is used to detect mechanical faults but in the rotor case, this technique can be used successfully because the broken rotor bar will excite the electromagnetic field disturbance which increases the torque modulations and hence lead to vibration which is easily measured by accelerometers (Kanovic et al., 2013).
The current will not flow in a broken rotor bar and the surrounding field will not exist. Because of that, the forces will be different from both sides of the rotor, unbalanced magnetic which rotates at rotational speed and modulates several poles times slip frequency will be created. So, the spectrum will have an increase in amplitude with sidebands at the rotational frequency (Kanovic et al., 2013). Eccentricity Faults occur due to an uneven air gap between the stator and rotor in the motor. The eccentricity faults are divided into 3 parts: Static, Dynamic, and Mixed eccentricity as shown in Figure 7. Considering the literature, eccentricity faults can be detected by vibration monitoring by observing sidebands concerning rotor slot frequency or by supply frequency (Ch et al., 2015). Figure 7. Eccentricity faults (a) normal (b) static (c) dynamic eccentricity (Gangsar & Tiwari, 2020) For analysis purposes, the motor will be in two states coupled and decoupled. Coupled motor will be seen as a whole system (motor + load) assembly and the decoupled motor will only consist of the motor side. From the literature (Ch et al., 2015), we can reliably say that current monitoring methods are more successful in monitoring eccentricity faults in decoupled motors on the motor side. But when we see it as a whole system, vibration monitoring is more successful because the fault can be on the load side and the current monitoring would not be able to detect the fault at an early stage.
Unbalancing Faults, in general, occur in rotating parts of a machine, e.g., rotor unbalance in an IM which happens when the center of mass does not coincide with the geometric center of the motor. The main causes are manufacturing defects, unwanted chipping or addition of mass on the rotor, thermal expansion, or bending of the shaft. Classified into 3 categories Static, Couple, and Dynamic unbalance. A centrifugal force is produced by the unbalancing due to which there are vibrations at a frequency equivalent to relative shaft speed and due to mutual inductances becoming unsymmetrical between stator and rotor the stator current harmonics occur at frequencies (Rahman & Uddin, 2017) calculated by (7) (Gugaliya et al., 2018).
is the unbalanced rotor frequency, is the electrical supply frequency, is per unit slip, is the number of poles.
Misalignment Faults are common faults like unbalance and occur when the coupled shaft center is not coinciding with each other as shown in Figure 8. In short term, it reduces the efficiency of the machine and in long term, it can also cause the failure of the machine. Flexible couplings are usually used to eliminate this fault. These are classified into parallel and angular misalignment. Vibration analysis and current analysis are used to detect the misalignment fault by observing harmonics, 3x will be highly excited as compared to 2x, 4x, and 5x harmonics of vibration and current (Kumar Verma et al., 2013).  (Gangsar & Tiwari, 2020) Cavitation is a general issue when it comes to pumps, the phenomenon happens when the water pressure drops below the threshold value which causes vaporization and formation of tiny bubbles, the bubbles create a shockwave while imploding and hence excessive vibration on the pump casing which is detectable. It causes low performance, damage to the impeller and volute, bearing failure, and seal failures so it is very important to detect and eliminate the fault at the incipient level Figure 9. Shows cavitation in centrifugal pumps (Dutta et al., 2018) (Stopa et al., 2014). Figure 9. Pitting due to cavitation in centrifugal pumps

Current Monitoring
Winding faults can happen due to heating, electrical, environmental, and mechanical stresses which affect the stator, the insulation breaks and causes a short circuit in the motor due to which the motor can heat excessively, high current flow, high voltage flow, physical damage, etc. The Park vector approach is a good method to determine winding failures. Equations shown below (8-12) (Ye & Wu, 2000) represent a circular locus centered at the origin of the coordinates. The Equations (11-12) will not be valid for any abnormalities in the motor.
Under ideal conditions, 3-phase current corresponds to Parks vector with these components , , are main phase variables, , are Park vectors current components, ' 'is the maximum value of supply phase current (A), is angular supply frequency (rad/sec), is the time variable. The diagnosis is based on the elliptical pattern which corresponds to the motor current parks vector form. Ellipticity changes and major axis orientation will tell fault and severity in the diagram (Ye & Wu, 2000).
Rotor bar faults will produce a rotor asymmetry which will lead to the resultant backward rotating field at the slip frequency respective to the forward rotating rotor, due to this backward rotating field concerning the rotor induces electromagnetic force and current in stator winding (Mehala & Dahiya, 2007). The sidebands can be detected at twice slip frequency (12) (Mehala & Dahiya, 2007) = (1 ± 2 ) is sideband frequency, is supply frequency, is slip Current monitoring techniques like Motor Current Signature Analysis (MCSA) are equipped with tools to detect rotor bar faults.
Eccentricity faults cause special patterns unique to the fault and can be detected by current spectrum analysis. The rotating wave approach method is used by which the magnetic flux waves in the air gap are calculated by magnetomotive force waves multiplied by permeance.
Bearing faults associated with components like cage, balls, inner race, an outer race are detected by their respective CFC which could be done by vibration or current spectrum, is used to diagnose bearing faults in the current spectrum (15) (Gugaliya et al., 2018) is the relative frequency between and , is supply frequency, is characteristic fault frequency, k=1, 2, 3...
Gearbox faults lead to failure of machines, malfunctions, and financial losses so it's essential to conduct CM and diagnosis of faults. The gears can have gear cracks or broken teeth that have to be detected. Vibration signals show modulations by output shaft rotating frequency, current signals are highly modulated by input shaft rotating frequency in case of gear cracks and two broken teeth. The current-based approach is more sensitive to low-frequency ranges and vibration is more sensitive to higher ranges, compared to vibration current monitoring is non-intrusive and less sensitive, so it has a high potential to be used in the commercial sector (Jin et al., 2016).

Thermography Monitoring
Infrared thermal imaging for motors and machines is a noninvasive method to detect faults that produce localized heat. Thermal image cameras like FLIR can be used efficiently to take a thermal image of a running motor or load and determine if there is an unusual or comparative difference in temperatures from the healthy motor thermal image temperatures.
Comparisons are done on the relative temperatures ∆ by (16) (Reljić et al., 2016) is the local temperature (interest point) is ambient temperature By ∆ we can determine the temperature rise in our point of interest on the machine. If the rise is severe, it is a sign that a fault is present at that location of temperature rise Figure  10. (Reljić et al., 2016). Shows the broken rotor bar fault with its effect on its neighboring bars and it is very clear what type of fault is visible Figure 11. (Choudhary et al., 2019). shows different types of bearing faults (a) lack of lubrication (b) inner race defect (c) outer race defect (d) healthy motor. The images can also be processed for better color resolution, as we can see in the images that all the faults have a different thermal image which can be used to detect and identify faults.

Acoustic based Monitoring
Acoustic is sound-based monitoring and fault diagnosis technique that can be done by low-cost capacity microphone-computer setup or digital voice recorder. For CM frequencies below 100Hz are essential so the sound recorder should be able to capture to work in low-frequency ranges. Data collection is very cheap in this method, the process is noninvasive, and instant collection of data can be done. A basic layout of the acoustic-based CM system is shown in Figure 13. Many faults like multiple broken rotor bars, the broken ring of squirrel cage, differentiating healthy with faulty IM, and more, with good accuracy, can be achieved by this type of system in addition to signal processing techniques (Glowacz, 2018). Figure 13. Acoustic monitoring setup The acoustic emission technique is also a good technique for detecting electrical and mechanical faults, it works better for electrical faults. The technique uses a load or stress which is generated at the machine and is detected by the sensor, the wave is then sent to the analyzing instrument to detect faults. A schematic diagram is shown in Figure 14.

Flux Monitoring
Flux monitoring works on the external magnetic field or leakage flux or stray flux, the magnetic flux density is used with stator current and compared in the frequency domain to detect different types of faults. Error! Reference source not found. Figure 15 and Error! Reference source not found. (Negrea, 2006) show the flux leakage and set up with the sensor (Soother & Daudpoto, 2019) (Negrea, 2006). Figure 15. Leakage flux (Negrea, 2006) Load/Stress Machine AE Sensor AE wave AE analyzer Flux monitoring is a good technique to find rotor cage faults with good accuracy proportional to motor loading. The short circuit in the stator winding is also detectable easily. Dynamic eccentricity can also be detected with circulating currents with ease. Rotor bar faults are seen but can be difficult to differentiate the signature with winding interturn fault (Goktas et al., 2017). Bearing fault and static eccentricity were not detectable by this method (Vitek et al., 2011;Negrea, 2006). Fault diagnosis capability, in terms of type and severity level of faults, of several CM techniques has been discussed in this section. The next section proposes a methodology to rationally select the most appropriate (combination of) CM technique (s) for a given industrial application.

SELECTION OF CM TECHNIQUES
For complete asset CM, it is important to consider both motors as well as the load-side faults. Every technique has its uniqueness to detect faults at an incipient or severe level. In this paper, a method is proposed to find which technique is the best for complete asset monitoring by considering many factors associated with real-time applications for every technique which is a major concern for industries and generally not considered by researchers.
The industries are moving closer to industry 4.0 by adopting new and better methods like remote monitoring, industrial IoT, and wireless data transfer to make data collection easier. Conventionally the engineer will go to the site and collect data from the data acquisition system (DAQ) system near the machine, by adopting new concepts and making things smarter and automated the sensors can send the data to the IoT gateway using serial communication or using Arduino chips as DAQ to collect data and then transfer to the cloud. From the cloud, we can get the data into our system by analyzing software that can diagnose the data continuously. If there is a fault it can be repaired before it has a significant impact on the working of the whole system Shyamala et al., 2017;Yaseen et al., 2017) Only a few papers have discussed the proper selection of CM techniques for rotating machinery, most of them only consider the cost aspect. Some researchers also focused on sewers and water mains applications. On a general basis  Table 3 compared to the proposed methodology. As the process involves extensive knowledge of machines, strategies, net present value, internal rate of return, faults, economic analysis, and many other important factors. Expert opinion for making a wise decision becomes a must. Hypothetical examples are given by researchers to deal with multiple variables and some case studies are presented to combine factors into an informed decision but still, the process lacks in the signal processing aspects related to sensor data, data collection, fault category combinations, and criticality of the asset. These factors are very critical to the industry and it needs to be addressed. The proposed methodology addresses these problems and overcomes the disadvantages of other methods as well. The proposed methodology uses the weighted sum model (WSM) of MCDM with an addition of a justification factor for fault categories. It is simple, easy, non-computational, less complex, includes many criteria, and has a certain outcome. This ensures that every industry can adopt and implement a maintenance strategy.

AHP Complexity increases with variables
The validity of the proposed methodology is generalized for rotating machinery having a load driven by a motor. As discussed in Section 1, the most common motor and loads have been considered from the literature for which data is presented in Figure 1 and Figure 2. The methodology can be generalized for other types of assets taking help from an expert. Most common faults are considered for both motor and load sides as discussed in Section 1 and Section 2.
The proposed methodology for finding the technique ranking per industrial factor is shown in Figure 17. The proposed methodology for the best technique combination including industrial implementation factors and different system requirements is shown in Figure 18.  A general rating scale of 3 criteria (0,1,2) is chosen. High scores are desirable and 0 is the worst. Details for each factor are given separately in their respective score distribution descriptions.

Score distribution description for matrices A and B
Scores are proposed as per the capability of the techniques to detect a fault at different severity levels. Table 4 shows the fault severity descriptions of all discussed faults. Table 5 and  Table 7. The best techniques with their next best alternative are chosen from Table 7 and presented in Table 8.  Table 9 is the motor fault detection priority distribution matrix, the score is distributed according to the criticality of the fault at the incipient level. Bearing and winding faults are the major issues that occur in motors, in which winding fault is considered more severe. It increases the stresses and causes temperature rise which according to a rule 10° rise in temperature reduces the life of insulation by half. If a winding fault happens it takes a considerable amount of time to repair and downtime is high increasing industrial losses (G. . Matrix D Table 10 contains the priority for load-side faults, in which pumps are mostly used. Cavitation causes erosion, implosion, misalignment, decrease flow, and greatly reduce efficiency, so it is very important to detect cavitation faults at the incipient level (Dutta et al., 2018;Stopa et al., 2014).

Score distribution description for matrices C and D:
The score is proposed according to the criticality of faults, fault priority is given high if it can have a preposterous effect if not detected at the incipient level. Matrices E, F, G: Table 11 is based on important factors which have to be considered while implementing the technique. Matrix E is the cost factor score, cost is considered a major concern for many industries when it comes to implementing and adopting a new system, because of the cost many industries don't even consider implementing CM. But for critical systems, it is a necessity, or it can have major implications like shutdown or production loss. Matrix F is how easily the data collection process is done by a technique that is necessary for saving the time of engineers. Matrix G is the noise factor, industries have a lot of machines running round the clock, and a lot of external noise affects the monitoring system, especially the acoustic sensors are very sensitive to external sound and so a less score is given to the technique.

Score distribution description for matrix E, F, G:
The scores are proposed according to the importance of factors considered by industries when selecting a CM technique. The higher the importance of the factor the higher the score is given.

Cost:
1: Equipment cost is relatively higher 2: Equipment cost is relatively lower Ease of data collection (EDC): 1: Data collection will require personnel to visit the machine 2: Data collection can be done from a control center Noise factor (NF): 1: Technique is very sensitive to background noise 2: The technique is not sensitive to background noise Table 11. Ease of application score matrix E, F, G (motor + load side)

Matrix operations:
The basic methodology adopted here is to get a scoring matrix dependent on faults themselves which is the basis of a technique selection. Then to get a single score for each technique, all the values in a row are added technique-wise. Now the weights of each factor can be multiplied by individual matrices for getting a final score of individual techniques.
[A]: Motor side fault detectability score matrix for common techniques.
[B]: Load side fault detectability score matrix for common techniques.
[C]: Motor side faults priority score matrix [D]: Load side faults priority score matrix [E]: Cost score matrix for common techniques.
[F]: Ease of data collection score matrix for common techniques.
[G]: Noise factor score matrix for common techniques.
[LFDFP] = Step 3: Weightage is assigned to each factor: WM: Fault detection and priority on the motor side.
WL: Fault detection and priority on the load side.
WC: Cost to implement the technique on the asset.
WD: Ease of data collection.
WN: Noise factor associated with a technique.
MCDM matrix is made using techniques, criteria, and weights assigned to all the criteria and is presented in Table  12. As all the values have a different range they are normalized as shown in Table 13.  Table 14. Both mechanical, as well as electrical faults, can happen in an industrial asset. As evident from Section 2, motors side electrical faults are prioritized. Loadside mechanical faults are significant. TCJS Technique combination justification score concerning combinations consisting of best techniques in detecting mechanical as well as electrical faults together. TCJS scores are presented in Table 15 with their normalized values.

Techniques TAS by MCDM
Vibration WM x 0.714 + WL x 1 + WC x 0.5 + WD x 0.5 + WN x 1 Step 4: The obtained TAS scores are added corresponding to technique combinations and then multiplied by their respective normalized values of TCJS to calculate the final TCAS values as shown in Table 16.

Score description for TCJS:
1: Technique combinations not consisting of best techniques from both mechanical and electrical faults category.
2: Technique combinations containing best or alternative best techniques one from mechanical and the other from the electrical category of faults. Refer to Table 8.

Current + Thermal
[TAS (Current) + TAS (Thermal)] x 0.5 Table 16. TCAS scores for technique combinations TCAS Score is based application of technique combination package for the complete system (motor + load) considering critical faults detection, ease of application, cost, ease of data collection, and noise sensitivity with taking into account the main focus of the implementing CM technique is to detect both mechanical as well as electrical faults. The location of sensors and how easily the data can be collected are taken into consideration. Sensors are combined in a solution like smart sensors packages for online monitoring of critical systems. It is easy to collect data from one location and better if the engineer does not have to visit the site for that purpose. All these factors are taken into consideration for making the final rankings. The expert team can be consulted to carefully choose the weights and find the best technique combinations for their asset. Each asset has its criticality and needs fault monitoring systems which can be best understood by the experts in that field.
There are different kinds of assets with different needs like equipment cost, space, and downtime cost, therefore generally a single technique is not sufficient to fulfill the needs of CM. To eliminate such shortcomings, a combination of sensors has to be used. Also evident from MCDM score values shown in Figure 20 to Figure 24, when independent techniques are compared to technique combinations. Assets can be classified into 3 different categories such as critical, sub-critical, and non-critical. All these systems have different factors which are considered by industries like cost, downtime, effectiveness, ease of implementation, complexity, external factors, shutdown cost, and maintenance strategy. (Bellini et al., 2008;Guoji et al., 2014;Henriquez et al., 2014;Schütze et al., 2018;Shin & Lee, 2015;Trajin et al., 2010;Uddin et al., 2014;Zhang et al., 2012). Figure 19. Example of system criticalities on an assembly line (Critical, Sub-critical, non-critical) Figure 19. Example of system criticalities on an assembly line (Critical, Sub-critical, non-critical) shows an example of different system criticalities. System 1 is referred to as a critical system because if it fails the whole system shuts down, so low downtime is essential and the best techniques have to be adopted. System 2 is sub-critical because it has a redundancy that can be used while the system is in maintenance but not for a long time. System 4 is non-critical because even if it fails the production is still smooth but can reduce efficiency, so cost-effective techniques have to be chosen.
According to literature and expert opinion the weights for critical, sub-critical, and non-critical assets in some industrial sectors like oil refinery, chemical plants, power generation, water applications, material handling, agriculture, manufacturing and packaging are mentioned in Table 17. In critical systems, the main focus is to protect the system and prevent any fatal failures which are responsible for shutdowns. 70 % of weightage is given to motor and load fault detection and fault priority criteria, other criteria are of lower priority in this type of system. Sub-critical systems are given 50 % weightage to fault detection and priority and 40 % to cost as these systems usually have redundancies that can operate for some time if these systems are under maintenance. A balance between cost and quality of CM is usually maintained. Non-critical systems can usually be shut down or can be replaced if failed, their failure has a very low effect on the whole system. Cost criteria have the highest weightage in these systems at 70%. (Al-Najjar, 1999, 2007, 2012Al-Najjar & Alsyouf, 2003;Maletič et al., 2015) With the given weights from expert opinion, TCAS scores of all the possible technique combinations were obtained for critical, sub-critical, and non-critical assets as shown in Table 18 to  Table 21. Only the best technique is recommended for critical assets. It is well evident from Section 2 that vibration and current are the best techniques for mechanical and electrical faults. Acoustic which is the best alternative for vibration made a pretty good combination with current and flux which was the best alternative for current made an appreciable combination with vibration to obtain the top 3 positions in sub-critical and non-critical assets.

Criticality Ranking Technique Combination
Critical  Figure 20 to Figure 24. To compare the combinations with independent techniques, TAS values (MCDM values) of both techniques are added and a percentage increase in the score is evaluated with respect to the base technique. For comparison, the case of critical assets is considered as having the most weightage towards the quality of fault detection.
As evident from the given chart technique combinations are more effective concerning rotating machinery where both mechanical, as well as electrical faults, are present with other criteria and difficulties.

Figure 20. Percentage improvement when vibration is combined with other techniques in critical assets
When a single technique is used for both motor and load sides only a single category of faults can be detected as evident from Table 5 and Table 6. For example, if vibration monitoring is used on both sides and the winding fault in the motor the whole system will be shut down without any fault indicator. The best combinations can be effectively used in industries for health monitoring of complete assets (motor + load). The combination of Vibration-acoustic was eliminated by careful observation that both detect similar faults. Industries that do not want to spend much on CM for non-critical assets, can adopt the recommended technique for specified industries or in similar applications as per expert opinion. Industries willing to spend some money on sub-critical assets can adopt recommended combinations for costeffective monitoring. For critical applications, the main objective is to reduce downtime, vibration-current combination which got the best TCAS score is recommended. Implementation of recommended techniques will surely help in the maintenance of rotating machinery while also saving a lot of time that goes towards unplanned maintenance. Costs associated with shutdowns and machinery costs can be saved with the recommended technique combinations for specific needs. Figure 25 shows a typical implementation system for CM of rotating machinery. A multi-sensor package will transfer the machine data to DAQ which then will be converted from an analog signal to digital, and directly send by serial communication. Binary signals can be stored in a large quantity in small databases and then can be sent using Industrial Internet of Things (IIOT) technology which will be uploaded to the cloud. Information will be downloaded and decrypted to its original form by DAQ software.
Original information like vibration data, temperature, sound, flux, and current which is recorded from MCC Panel (Machine Control Centre) will be analyzed by the Fault Detection Model. If there is any unusual behavior or fault it will be immediately reported to the maintenance engineer, and steps will be taken to correct the problem.

CONCLUSION
It is evident from the techniques and scores that the best techniques of CM are vibration for mechanical faults diagnosis and current for electrical faults diagnosis. But it is not practical to implement these in every scenario. Industries consider cost and other factors into account for diagnostic setup and if the diagnostic system will cost more than the asset cost, the industry will not even implement CM in the first place. This paper tries to solve that problem by giving them the flexibility to choose a suitable combination of techniques for different levels of sophistication. As per industrial needs expert opinion could be taken for deciding the weights for given criteria and final scores can be obtained from the mentioned methodology. A set of weights were obtained from industrial experts for critical, sub-critical, and non-critical assets in some industries. Recommended combinations can be used by industries and in similar applications. Factors like cost, noise, criticality, multiple fault category justification, and ease of data collection are already considered in the methodology so that the results can be directly applied to real-time rotating machinery. A comparison is also shown when a technique is implemented independently and when it is combined with other techniques to help understand the advantages of combining techniques. Also, a suitable wireless setup is suggested, considering the latest advancements in technology for remote monitoring of rotating machinery.