Deep learning-based fault diagnosis systems for prognostics and health management of mechanical systems is an active research topic. Notably, the absence and class imbalance of fault data (insufficient fault data compared to normal data) have been shown to cause many challenges in developing fault diagnosis systems for the manufacturing fields. Therefore, this paper presents case studies using deep learning algorithms in the absence or class imbalance of fault data. Auto-encoder-based anomaly detection method, which can be used when fault data is absent, was applied to diagnose faults in a robotic spot welding process. The anomaly detection threshold was set based on the reconstruction error of trained normal data and the confidence level of the distribution of normal data. The anomaly detection performance of the auto-encoder was verified using non-trained normal data and three sets of fault data through the threshold. As a case study for insufficient fault data, synthetic data was generated based on cGAN and applied to diagnose fault of bearing. Using the imbalanced dataset to generate synthetic fault data and to reduce the imbalance ratio, it was confirmed that the accuracy of the synthetic data generation-based 2DCNN fault diagnosis model was improved.

This paper proposes a high-fidelity finite element model of a permanent synchronous motor (PMSM) to predict electromagnetic responses. The proposed method aims to generate electromagnetic responses from the PMSM under various operational conditions-including normal and faulty conditions-by coupling several partial differential equations governing the electromagnetics of a PMSM. The rotor eccentricity is considered to be a representative fault of a PMSM, which has electromagnetic characteristics that differ from the healthy state of a PMSM. Note that eccentricity is the most frequent fault during PMSM operation. Therefore, the proposed model could replicate the defected torque responses of an actual motor system. The effectiveness of the proposed model is validated using measurements from a PMSM test bench. Quantitative comparison reveals that the proposed model could replicate both the transient- and steady-state torque responses of the PMSM of interest at a variety of operational conditions, including a faulty status. The proposed model could be used to generate virtual electromagnetic responses of a PMSM, which could be used for data-driven fault detection methods of electric motor systems.

A hairpin motor is a type of motor that is used for driving an eco-friendly car. Unlike a conventional coil-winding motor, hundreds of hairpins formed by an enameled copper wire with a rectangular cross section comprise a stator to improve the driving efficiency by maximizing a coil drip rate. With the increased use of the hairpin motor, there has been an increased interest in manufacturing techniques and automated systems of the hairpin motor. Enamel coating removal is one of the major processes of hairpin motor production; enamel coating at the end of the hairpin should be removed to connect the hundreds of hairpins by using the welding process. Grinding is one of the machining processes used for removing the enamel coating. This study proposed an adaptive control method for the grinding process to improve the efficiency and quality of the enamel coating removal process. Grinding depth is maintained during machining by controlling the vertical position of the spindle based on driving torque. A lab-scale grinding machine including a sensory system for adaptive control is developed and used to verify the performance of the proposed method.

This research developed a CAM S/W, which generates an adaptive 5-axis tool path, to optimize the quality of Direct Energy Deposition (DED) 3D printing. After reconstructing part shapes and generating printing paths in each shape, the path simulation including automatic collision detection was implemented. Productivity and printing quality were improved through equipment improvement and process optimization. In addition, high-quality parts with desirable physical and mechanical properties were produced by generating an adaptive 5-axis path specialized in the printing process that reflects various physical phenomena and monitoring results. Finally, the performance of CAM S/W was verified through the production of prototypes for industrial components.

Currently digital transformation has a huge impact on human lives. Digital transformation does not just mean a transformation of a (non-) physical element to a digitally identifiable element. It focuses on the utilization of digital technology for transforming (improving) procedures or routines of business and operation. The manufacturing industry has been adopting the most recent digital technology, and lots of digital data are being created. To utilize the stored data, data analysis is essential. Because the manufacturing data is created in a different format at every manufacturing step, the integration of the data is always the bottleneck of the data analysis. Querying of the right data at the proper time is fundamental for high-level data analysis. The digital thread is introduced to provide the inter-reference of digital data based on a context. This paper proposes a digital thread framework for the machining process. The context of the proposed framework consists of the questions of how the product will be machined, how it is (was) being produced, and how it was made. A prototype software was developed to verify the proposed framework by implementing the creating, storing, and querying modules for simulation, monitoring, and inspection data.

Drone is an innovative industry that can combine the application of various technologies in the fourth industrial era, such as big data, artificial intelligence, and ICT. Although the synergy effects of these technologies will be great in various industrial ecosystems, drones are vulnerable to gusts such as "building wind" or "valley wind". Herein, the frequency domain of a mini drone was identified and a model-based disturbance observer (DOBs) was applied to implement the drone robust resistance against gusts. The frequency response of the Parrot Mambo or mini drone was measured with multi-sine excitation and the system dynamic parameters were identified. Based on the identified model, DOBs were designed and applied to the drone’s altitude, position, and yaw control. The effectiveness of the DOBs was verified with a sinusoidal disturbance. With the model-based DOB, 84.5% of the drone altitude responses, 50.7% of x responses, 52.1% of y responses, and 79.7% of yaw responses against sinusoidal disturbances were reduced. Flight responses were measured against wind disturbances with changing speed and direction. With the model-based DOBs, the drone"s altitude decreased by 87.7%, the x position by 53.0%, the y position by 60.6%, and the yaw angle by 56.2%.

M.2 NVMe SSD (Non-Volatile Memory express Solid-State Drive), which have higher computational speed and reliability than conventional devices, have come to be widely used. Recent studies have reported that M.2 NVMe SSD are beginning to have thermal issues due to the increasing heat generation occurring with the high chip density and high-performance operation in a limited space. Thermal issues in the controller and memory units of M.2 NVMe SSD lead to increased failure rates and decreased data retention times. In this study, we propose a compact and optimized thermal solution for commercial M.2 NVMe SSD installed between the mainboard and GPU (Graphic Processing Unit). A thermal and fluid dynamics simulation of an M.2 NVMe SSD, including the heatsink, was performed, and the Genetic Algorithm method was used to optimize the heatsink size.

This study performed high-frequency heat treatment experiments and simulations of the park gear of an automobile transmission. The heating temperature and hardening depth were measured during high-frequency heat treatment. Moreover, by applying the resonance RCL circuit, the current value of the coil during high-frequency heat treatment, the electromagnetic and heat transfer material properties dependent on the temperature, and the phase transformation function were all applied to the simulation. In the high-frequency heat treatment experiment, the heating temperature was 977.4℃ and the 1st direction hardening depth was 1.5 mm, the 2nd direction hardening depth was 3 mm, and the 3rd direction hardening depth was 2.5 mm, and the reliability was verified by comparing the simulation heating temperature of 1,097℃ and the 1st direction predicted hardening depth of 1.6 mm, the 2nd direction predicted hardening depth of 2.8 mm, and the 3rd direction predicted hardening depth of 2.7 mm. The error rate of the heating temperature results was 12.2% whereas that of the hardening depth results was 7.1%.

The WC-5wt.% TiC compacts, which was fabricated by pulsed current activated sintering process (PCAS), were cryogenically treated to improve the mechanical performance. The densely consolidated specimens were exposed to liquid nitrogen for 6, 12, and 24 h. All cryogenically treated samples exhibited compressive stress in the sintered body compared with the untreated sample. The cryogenically treated samples exhibited significant improvement in mechanical properties, with a 9% increase in Vickers hardness and a 52.6% decrease in the fracture toughness compared with the untreated samples. However, excessive treatment of over 12 h deteriorates the mechanical properties due to tensile stress in the specimens. Therefore, the cryogenic treatment time should be controlled precisely to obtain mechanically enhanced hard materials.

In this paper, we describe the development of a 5-axis force/moment sensor of an intelligent gripper designed to grasp the weight of an unknown object and the position of the object in the gripper. The 5-axis force/moment sensor consists of an Fx force sensor, Fy force sensor, and Fz force sensor to measure weight, along with an Mx moment sensor and Mz moment sensor to determine the position of an object in the gripper. These sensors are all built within a single body. Each sensor sensing part of the 5-axis force/moment sensor was newly modeled and custom designed using software, and each sensor was manufactured by attaching a strain gauge. The results of the characteristic test of the fabricated 5-axis force/moment sensor showed that the rated output error was within 0.1%, the reproducibility error was within 0.05%, and the nonlinearity error was within 0.04%. Therefore, the 5-axis force/moment sensor developed in this paper can be attached to an intelligent gripper and be used to grasp the weight of an unknown object as well as the position of the object in the gripper.