Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
MACHINE LEARNING-BASED PREDICTIVE METHODS FOR. POLYPHASE MOTOR CONDITION MONITORING by. David Mathew LeClerc. A Thesis. Submitted to the Faculty of Purdue ...
Typology: Schemes and Mind Maps
1 / 51
This page cannot be seen from the preview
Don't miss anything!
by David Mathew LeClerc A Thesis Submitted to the Faculty of Purdue University In Partial Fulfillment of the Requirements for the degree of Master of Science Department of Engineering Technology West Lafayette, Indiana August 2022
Dr. Grant Richards, Chair School of Engineering Technology Dr. Gaurav Nanda School of Engineering Technology Dr. Kenneth Burbank School of Engineering Technology Approved by: Dr. Duane D. Dunlap
I would like to thank my advisor, Dr. Grant Richards, for pushing me and helping me throughout my Graduate program. I would also like to thank Dr. Ken Burbank for supporting me throughout my program. I want to thank Dr. Gaurav Nanda for his machine learning expertise and patience. Lastly, I want to thank Professor Dale Nicholson for supplying me with sensors and for taking the time to give me valuable insight. Thank you to Fluke, Frederic Boudart, Michael Stuart, and Kevin Clark for helping with this thesis project. The technical support, as well as the supply of tools, was crucial for the project. I want to thank Kirby Risk, Marty Guy, Travis Stafford, and Jackson Maish for supplying the test motor and assisting with motor knowledge.
Artificial Intelligence (AI) Preventive Maintenance (PM) Predictive Maintenance (PdM) Condition Monitoring (CM) Industry 4.0 (I4) Overall Equipment Effectiveness (OEE) Internet of Things (IoT) Critical Maintenance Event (CME) Sequential Minimal Optimization (SMO) Machine Learning (ML)
1.1 Background Distribution centers are crucial for supply chains in every industry and are the heart of modern commerce systems. The heart of a distribution is the conveyance system, allowing for the efficient movement of goods in all phases from intake, sort, storage, picking, and packaging. Conveyor systems incorporate many physical assets such as motors, gearboxes, chains and sprockets, belts, rollers, compressed air, and bearings. Distributors face increased demand for goods and expectations for decreased time to deliver goods. Standard solutions to address these challenges include advanced conveyor systems, with modern distribution centers commonly having miles of conveyor systems in a single site. Newer extended and advanced versions of traditional conveyor systems require more motors than previous conveyance systems. One example is Amazon, the largest distributor of manufactured goods in the United States, grossing over $152 billion in 2020 (Macrotrends, 2021). Amazon uses a complex network of over 110 distribution centers called "fulfillment centers" in the United States alone (FBA Help, 2021). Each Amazon fulfillment center has the technology to sort, store, and transport millions of dollars of product each hour. Conveyor technologies are the most critical aspect of the centers, with one Amazon fulfillment center in Gaines, Michigan, having 22 miles of conveyor system (Linnert, 2021). Historically the maintenance strategy employed in distribution centers was a run-to-failure maintenance strategy that reacts to an asset failure, triggering a repair or replacement. Running an asset to failure is not the most effective maintenance method for conveyor systems. The run- to-failure maintenance method can produce hours of downtime, costing a manufacturer or distributor millions of dollars per hour. United States manufacturers are another group heavily utilizing polyphase motors and are familiar with unplanned maintenance costs. "Unplanned downtime costs industrial manufacturers an estimated $50 billion annually" (Moyle, 2021). Technologies capable of making run-to-failure maintenance a method of the pas can be found under the scope of Industry 4.0. The newest wave of technology brings sensors, maintenance software, and diagnostic tools to prevent and even predict an asset's maintenance
needs. New technology such as MEMS microphones and acoustic imaging allows issues to be seen or heard before humans. These new technologies improve condition monitoring (CM) and condition monitoring systems (CMS). Condition monitoring monitors an asset's health to detect issues and promote longevity. According to Iqbal et al. (2020, p. 7), "the health of the equipment or process is assessed by collecting important and critical data from sensors and instrumentation." Modern CMS deployments can use AI technologies to help determine the deviation from a known performance metric at a given system state. This variation can be correlated with time to assess the time interval between a known state and the present state. 1.2 Research Problem Statement This work addresses the problem of the inability to predict motor maintenance accurately, as well as failure being time-consuming and expensive. The problem arises in distribution centers when motors are under-maintained or overloaded. The downtime from a failed motor will prohibit the user from making or delivering goods, potentially causing loss of production, reputation, or sales due to late deliveries. 1.3 Research Questions The specific research questions examined are:
2.1 Background The review of literature focused on initial research and advanced research techniques. The initial and advanced research was performed using search engines and databases, of which the most promising were ProQuest – Dissertations & Theses, google scholar, and science direct / Elsevier. ProQuest – Dissertations & Theses is a promising database to research work that has been done academically. Google Scholar and Science Direct are promising due to the amount of peer-reviewed, international journals and research articles in the respective directories. The ProQuest search results can be found below in Figure 1. The initial searches on ProQuest that did not include polyphase motors yielded hundreds of thousands of results. However, including "polyphase motor" shrank the hits to 48, 27, and 205. Including "polyphase motor" helped refine the search, so the search was proper. The original search query was changed to include a "polyphase motor" to refine the search and find relevant, helpful information. The updated search queries were used for all three databases. Figure 1. ProQuest Search Results The Google Scholar search results can be found below in Figure 2. Google Scholar searches needed to have some phrases in parenthesis to refine the search. Predictive maintenance needed to be in parenthesis, or the search included results for routine motor maintenance. Condition monitoring was placed in parenthesis as the complete phrase was left out of most searches. The Science Direct search results appear in Figure 3.
Figure 2. Google Scholar Search Results The Science Direct search results can be found below in Figure 3. The Science Direct search terms proved less valuable than other databases. The search query is too specific in searches 3 and 4. Searches 1, 2, and 5 yielded useful searches; however, Science Direct includes unhelpful chapters and bibliographies from irrelevant material. Figure 3. Science Direct Search Results Figure 4, found on page 15, is a concept map of the project terminology. The terminology process started with Industry 4.0 and then to AI and ML. Machine Learning models and condition monitoring systems were the next step. ML models and CMS’s enable predictive maintenance. The polyphase motor was the device that was analyzed and terms such as operation and reliability were analyzed to better craft the experiment.
the rotating flux wave, produced by the stator winding, and the rotating rotor- current wave, developed in the rotor winding, are traveling at the same speed as seen from a point on the stator. (p. 2) Figure 5. Nikola Tesla's Original Motor Design from 1889 The polyphase motor that Cochran describes appears in Figures 6 and 7. Figure 6 (Kuphaldt, 2007, p. 444) shows both two (a) and three (b) phase motors and the energized windings. Figure 7 (Kuphaldt, 2007, p. 449) shows the three-phase motor sinusoidal waves and the coils at half and full speed. Kuphaldt (2007) writes: For 60 Hz power, the magnetic field rotates at 60 times per second or 3600 RPM. For 50 Hz power, the magnetic field rotates at 50 times per second or 3000 RPM. Representing the synchronous portion of the motor, thusly (Kuphaldt, 2007, p. 449).
Figure 6. Two and Three Phase Motors with Energized Coils Figure 7. Full and Half Speed Motors with Corresponding Sinusoidal Waves The polyphase AC motor plays a critical role in industrial conveyor systems. DC motors require a DC power supply which costs more than AC. The novelty of an AC motor is to be less expensive and to take less power to operate. The AC motor is driven by a variable frequency drive (VFD), and the motor's purpose is to transfer the energy from the VFD to the conveyor system. The motor transfers the power to a coupling, pulley, flat belt, v-belt, sprocket and chain, gears, or a combination. The power transference method depends on the type of conveyor system; however, the two most common conveyor types are belt and roller. The belt conveyor systems use pulleys as the power transfer method, and the roller conveyor system uses a chain and sprocket method. The AC power supply is more cost-efficient and energy-efficient than a
2.6 Preventive Maintenance "Work performed as part of a fixed interval replace, repair or restore maintenance strategy" (Brundage et al., 2021). Preventive Maintenance (PM) methods include preventive measures to ensure failure does not occur. Preventive maintenance can be scheduled or routine actions such as a daily walk-through. Preventive maintenance methods also include activities such as regularly replacing parts. 2.7 Predictive Maintenance "A comprehensive predictive maintenance system uses the most cost-effective tools to obtain the operating condition of plant systems" (Mobley, 2002). Companies that use predictive maintenance (PdM) will also use a condition monitoring system (CMS). The CMS monitors asset conditions for the predictive algorithms in place. The software is taught the trends and will send alerts when maintenance or failure modes are detected. Mobley (2002) states that predictive maintenance using vibration analysis is possible because there are two constants." All failure modes have distinct frequencies that can be isolated and identified. The amplitude of each particular frequency will remain constant unless the asset's operating dynamics change" (Mobley, 2002). Twenty years ago, predictive maintenance was in the preliminary stages. Some companies and individuals attempted to define predictive maintenance. Definitions ranged from what is now called preventive maintenance to accurate predictive, condition-based maintenance. Mobley's (2002) definition of predictive maintenance is: To some workers, predictive maintenance is monitoring the vibration of rotating machinery to detect incipient problems and to prevent catastrophic failure. To others, predictive maintenance is monitoring the infrared image of electrical switchgear, motors, and other electrical equipment to detect developing problems. The common premise of predictive maintenance is that regular monitoring of the actual mechanical condition, operating efficiency, and other indicators of the operating condition of machine-trains and process systems will provide the data required to ensure the maximum interval between repairs and minimize the number and cost of unscheduled outages by machine failure. (p. 4) Four years before Mobley, Parrondo et al. (1998) described predictive maintenance systems based on continuous and periodic condition monitoring (p. 198). Parrondo et al. (1998)
explain that predictive maintenance techniques had been developing for the last few decades, and better techniques were being developed. “Reliable techniques, mostly based on spectral signal analysis, are now available for the detection of typical perturbation sources such as shaft misalignment" (Parrondo et al., 1998, p. 198). Beebe (2004) also describes predictive maintenance as condition-based maintenance. "Condition-based maintenance, also called predictive maintenance, applies to over 80% of maintenance" (Beebe, 2004, p. 4).
2. 8 Predictive Maintenance Technologies Technologies introduced under the scope of Industry 4.0 make predictive maintenance a reality. Industry 4.0 has brought about intelligent sensors, platforms, and ideas like the Internet of Things, artificial intelligence, and machine learning. 2.8.1 Smart Sensors Eifert et al. (2019) define intelligent sensors as "a multi-component measuring device that is self-calibrating self-optimized, and easy to integrate into the environment for high connectivity" (Bednar et al., 2021, p. 3; Eifert et al., 2019). Smart sensors are utilized in condition monitoring systems integrated into the Internet of Things. The Internet of Things (IoT) is the idea of interconnecting every asset, sensor, and computer inside an industrial environment. The devices communicate amongst themselves and alert human personnel if an issue arises. The smart sensors allow the technology to monitor the asset conditions. The sensors continuously monitor asset performance, and the data is constantly fed into a program to be modeled. The software will be taught what trends indicate maintenance and failure modes. The software can also alert human personnel if the maintenance and failure modes are detected. The Internet of Things will allow other assets to become aware of the alerts and adjust accordingly. 2.8.2 Internet of Things The Internet of Things allows for improved connectivity between industrial networks, assets, sensors, and computers. According to Michael Short and John Twiddle (2019), "of the many advantages leveraged by industry 4.0, the promise of increased integration and real-time
