1. Housing and Rotor Cage Production

Production starts from casing fabrication; however, no direct solution for the casting of motor casings has been found. During the casting process, the voltage or temperature can be collected and stored by sensors. Thereafter, machine learning technology in combination with artificial intelligence can be used to optimize the parameters in the casting process, predict the casting quality, and determine the appropriate casting material variables.

In addition, simulation methods can be employed to model the solidification of the liquid metal in the mold. The core can then be assembled by subordinate robotic arms, thereby reducing manpower and the probability of error. The asynchronous motor's rotor is manufactured by die-casting, which is also a method applicable to casing fabrication. Note that if the quantity of rotors to be manufactured is small, then it is necessary to prepare copper rods to be welded to the short-circuit ring.

#### 2. Laminated Core Production

The data collected from the laminated core manufacturing process using machine learning algorithms can be utilized to predict the quality of raw materials required for production. In the case of electrical board production, machine learning algorithms can classify the materials for electrical steel plates based on the electromagnetic properties of microstructures. During the cutting of electrical boards, by monitoring the cutting force, audio signals, or vibration signals through sensors, machine learning algorithms can be employed to detect and predict errors and build models. Virtual reality smart devices can also be used to assist in the cutting and welding operations. Using different techniques to cut out electrical boards, sensors can be used to monitor the punching forces, audio signals, or vibration signals. For the laser-cutting process, data mining through the use of machine learning algorithms can afford the potential of detecting errors and predicting key events. The quality inspection of cuts and the entire laminated core can be performed through computer vision.

#### 3. Insulation and Impregnation

In the motor manufacturing process, many of the steps involve providing insulation to motor parts. There are different alternatives for insulating the raw stator materials used in the slot, including the application of insulating paper powder coating or injection-molded polymers. Machine learning may be employed to aid in modeling this process and determining the best process parameters. Big data can also be combined to determine quality-related parameters and generate predictive models. After winding, the entire coil is insulated. In this aspect, machine learning can also identify the appropriate impregnation process simulation support.

#### 4. Winding

The technology of Industry 4.0 may be employed to set the sensor in the winding machine and guide the machine in creating a precisely positioned coil. Rodriguez et al. [11] proposed that a machine learning system can be used to optimize the wire contour generated by an automatic winding machine so that the coil contour is wound as uniformly as possible. Apart from accurate and non-destructive winding, it is also important to maintain the required coil resistance during enameled winding. This can also be optimized by machine learning by simulating the winding parameters (e.g., wire tension and winding speed). Combined with the image recognition technology, the use of a robot to wind and install coils can facilitate production and detect the cause of failure at any time.

#### 5. Contact technology

The existing method relative to contact technology directly solves the problem of crimping in motor production. It was reported in [12] that a standard OPC Unified Architecture condition monitoring system may be used to track the thermal crimping process (OPC Please See Table A1). In addition, Mayr et al. [13] studied the application of machine learning algorithms in the ultrasonic field for crimping. First, quality indicators (e.g., resistance and extraction force) can be estimated based on process parameters. Second, visual or auditory features can be used to classify joint quality.

In quality management, machine learning-based models are used as quality estimators, eliminating the necessity of quality management measures, such as random checks. For example, convolutional neural networks use visual features to predict joint quality. A comparison between deterministic models and machine learning methods shows that machine learning technology is more powerful, easier to automate, and more accurate. It can also detail the machine learning-based process control methods for the real-time measurement of parameters in the near future [14].

In addition to crimping, the potential of using machine learning in the laser welding of hairpin windings is studied. The application of laser welding contacts is particularly suitable for hairpin windings because of the large number of contact points. The use of machine learning can predict the quality of weld processing based on machine parameters. For the subsequent quality assessment, the combined image data can be used to detect and classify weld defects based on their severity. In the future, these applications will be incorporated into a quality monitoring system, which may also contain data from the upstream process. In particular, the burrs formed during the cutting process and residues resulting from peeling considerably influence the welding results [15].

#### 6. Shaft Production

Al-Zub Ardi [16] indicated that machine learning can be employed for quality management, process control, and predictive maintenance in the cutting process. The proposed model focuses on the prediction of surface roughness and cutting forces as well as the estimation of tool life and wear. In addition, the application of machine learning technology combined with edge and cloud computing allows the analysis of large data streams. Computer vision can be used for process monitoring or quality inspection of machined parts through machine learning algorithms. Different methods have also demonstrated that the OPC Unified Architecture condition monitoring system can aid in optimizing machining processes [17,18]. Moreover, compared with traditional machining centers, robots are convenient for machining operations because of their flexibility and relatively low investment costs. Finally, weight reduction can be achieved through the AM (Additive Manufacturing) of the shaft [19].

#### 7. Permanent Magnet Rotor Production

Various methods are employed in the production of motors, including permanent magnet synchronous motor rotors and those used in related fields. For example, image recognition can be employed for the visual inspection of magnetic surfaces after fabrication. In addition, sensors can be used for magnetic field measurements, which can be utilized to test a single magnet or the entire rotor. Magnetic field measurements also provide the basis for selective magnet assembly. Magnetic deviation has a critical influence on the operating characteristics of the motor, and it can be applied to optimize the deviation compensation magnet arrangement. Apart from traditional algorithms, machine learning techniques can be employed to develop optimal magnet assembly strategies. For each batch of magnets, a magnet or magnet stack is selected and installed according to an algorithm to minimize the deviation from the simulated magnetic field of an ideal rotor [20–22].

#### 8. Final Assembly and Testing

In the final motor assembly, various connection processes are involved, including press-in operations, gluing, shrinking, tightening, and welding. Data mining methods can be utilized to study the relationship between force–displacement and current curves in a press operation. The testing process naturally generates a considerable amount of data; hence, data-driven methods, such as machine learning, can be used to check multiple areas. For stator testing, machine learning methods can be used to evaluate and classify the fault mechanisms in electrical insulation. The quality characteristics measured herein can also be used as labels for predictive models based on machine learning that utilize parameters from previous processes, such as winding, insulation, and contact. In addition, several methods directly solve the end-of-line test of the motor, e.g., image recognition technology can support stator inspection. Machine learning technology capabilities can also be utilized for evaluating vibration signals [3] as well as acoustic data to analyze motors [23] and produce automatic fault detection functions.

#### 9. Overall Process

Apart from the application scenarios of Industry 4.0 technology in each sub-process, there are other methods related to the overall process that can include various steps in the value chain. For instance, with regard to the development of motors and related production systems, semantic technology can considerably improve the cross-domain information exchange. In addition to pure knowledge management, from the perspective of a configurator, knowledge-based systems can be used to automate simple engineering tasks. Thus far, it can only be found in motor design; however, it can also be employed in the engineering design of related production systems. For example, simulations can aid in analyzing the cost–benefit ratio of alternative production technologies.

With regard to production, knowledge-based systems can aid in optimizing individual jobs, and machine learning algorithms will be useful in processing big data and making fault predictions for the entire production line. Data mining can therefore facilitate the detection and rapid response to deviations within the assembly line. In this case, machine-to-machine communication technology performs an important function. For example, wireless communication technologies, such as Bluetooth 4.0 can be used to identify and locate tools and objects [24]. In the human–machine interface field, virtual reality-based auxiliary systems can support complex assembly processes [25]. In motor production, simple tasks can be handled with sensitive lightweight robots and automated guided vehicles. Another approach involves developing machine learning-based controllers for robots that can reduce the programming effort for assembly tasks [26]. In this case, machine or robot control can be transferred to the cloud and be provided as a service [27].

With the foregoing, the relationship between the motor production process and Industry 4.0 is established. Moreover, the application of the motor production process of the research object is summarized in Table 1.

In summary, the literature summarizes the intelligent systems that can be used in motor manufacturing, and also supplements the system that is lacking in existing literature. For the industry, it also provides a practical intelligent system model.


**Table 1.** Comparison table of motor production process and Industry 4.0.

#### **2. Materials and Methods**

#### *2.1. Research Architecture and Methods*

In this study, secondary data collection, field study, and interviews are performed to explore the selection of intelligent systems to be applied to Solen company, which is Vietnam's motor production line. In terms of secondary data collection, this study has searched for related research papers, journals, and reports on existing smart manufacturing cases; production line standard operating procedure manuals are also obtained. From these, feasible methods for "SM" motor production line are initially evaluated.

The main author and coauthor of this study also visited the Solen company Ho Chi Minh City factory in Vietnam on 19–22 September 2019 to conduct a field survey. This allowed the authors to observe the actual operation of the motor production line staff and understand each step of the operations. Based on this field visit, the processes and systems that smart systems may import from the technology of Industry 4.0 are initially assessed. (See Table 2)


#### **Table 2.** Research methods and data types.

A three-stage interview is conducted to collect expert opinions. 1. On 27 November 2019, two practical experts were invited to the laboratory for interview. 2. In order to understand the current situation of the company's production line automation, three persons in charge of the motor production line in the Taiwan headquarters of Solen company were interviewed on 31 January 2020 and from 14 February 2020. They were interviewed regarding feasible solutions for preliminary evaluation as well as the company's views on the introduction of intelligent systems and future planning. To ensure the correctness of the information, the consent of the respondents to record and photograph the interviews was obtained. It is also treated anonymously after the interview to protect respondents.

In this research, secondary data collection, field study, and expert interviews are mainly conducted to study the "SM" motor production line in Vietnam. Based on the observations of researchers of the actual production line and personnel interviews to understand existing production process, the smart machine tools (smart systems) necessary for smart manufacturing are introduced. Finally, the intelligence of the motor and coil assembly processes of this production line is analyzed. The research architecture is shown in Figure 1. In this study, there are three stages of research steps, Step 1: Secondary data collection, to widely collect data on the "SM" production line. Step 2: Field study, go to the production line for research. Step 3: Interview, this study conducted a total of 3 interviews to obtain a prosperity of research data.

**Figure 1.** Research architecture diagram.
