In machining operations, dynamometers are typically used to directly measure the forces acting on cutting tools. However, their high cost and complex setup restrict their use to laboratory environments, making them unsuitable for real-time monitoring in general production settings. To overcome this limitation, this study proposes an autoencoder-based learning model for estimating cutting forces using only spindle vibration signals acquired during milling. The model features a deep neural network (DNN) that takes processed spindle vibration signals as input and predicts latent features derived from cutting force signals through an autoencoder. These predicted latent features are then fed into a pretrained decoder to reconstruct the corresponding cutting force signals. To enhance the model's accuracy and robustness, the raw vibration signals sampled at 20 kHz were filtered with a bandpass filter that spans the effective frequency range of 20–2500 Hz, effectively removing irrelevant noise. For validation, an accelerometer was mounted on the spindle head of a milling machine, and vibration data were collected during cutting. The estimated cutting forces were compared to ground truth measurements obtained from a dynamometer. The model achieved a Pearson correlation coefficient of 0.943, demonstrating that reliable cutting force estimation is achievable using only low-cost vibration sensors.

Estimating energy expenditure is essential in monitoring the intensity of physical activity and health status. Energy expenditure can be estimated based on wearable sensors such as inertial measurement unit (IMU). While a variety of methods have been developed to estimate energy expenditure during day-to-day activities, their performances have not been thoroughly evaluated under walking conditions according to various speeds and inclines. This study investigated IMU-based neural network models for energy expenditure estimation under various walking conditions and comparatively analyzed their performances in terms of sensor attachment locations and training/testing datasets. In this study, two neural network models were selected based on a previous study (Slade et al., 2019): (M1) a multilayer perceptron using sensor signals during each gait cycle, and (M2) a recurrent neural network using sensor signal sequences of a fixed window size. The results revealed the following: (i) the performance of the foot attachment model was the best among the five sensor attachment locations (0.89 W/kg for M1 and 1.14 W/kg for M2); and (ii) although the performance of M1 was superior to that of M2, M1 requires accurate gait detection for data segmentation by each stride, which hinders the usefulness of M2.

Tire-related crashes account for a large proportion of all types of car accidents. The causes of tire-related accidents are inappropriate tire temperature, pressure, and wear. Although temperature and pressure can be monitored easily with TPMS, there exists no system to monitor tire wear regularly. This paper proposes a system that can estimate tire wear using a 3-axis accelerometer attached to the tread inside the tire. This system utilizes axial acceleration, extracts feature from data acquired with the accelerometer and estimates tire wear by feature classification using machine learning. In particular, the proposed tire wear estimation method is designed to estimate tread depth in four types (7, 5.6, 4.2, and 1.4 mm) at speeds of 40, 50, and 60 kmph. Based on the data obtained during several runs on a test track, it has been found that this system can estimate the tread depth with reasonable accuracy.

The tower crane is widely used in construction and transportation engineering. To improve working efficiency and safety, input shaping methods have been applied. Input shaping is a method of reducing residual vibration of flexible systems by convolving a sequence of impulses with unit step command. However, input shaping is based on the linear system theory in which its control performances are degraded, in case of nonlinearity and unmatched dynamics of the control systems. In this paper, a new optimal reference input shape design method based on minimizing cost function is suggested and applied, to a simple cart-pendulum system which is a simplified model of tower cranes. Since pendulum dynamics is nonlinear, analytic solution does not exist. To overcome this problem, in this paper, a machine learning approach is suggested to find optimal reference input shape for the cart position control. The feasibility of the proposed design method is verified through some simulation examples by using MatLab.

Recently, improvement of productivity of the paper cup forming machine has being conducted by increasing manufacturing speed. However, rapid manufacturing speed imposes high load on cams and cam followers. It accelerates wear and cracking, and increases paper cup failure. In this study, a failure diagnosis algorithm was suggested using vibration data measured from cam driving parts. Among various paper cup forming processes, a test bed imitating the bottom paper attaching process was manufactured. Accelerometers were installed on the test bed to collect data. To diagnose failure from measured data, the K-NN (K-Nearest Neighbor) classifier was used. To find a decision boundary between normal and abnormal state, learning data were collected from normal and abnormal state, and normal and abnormal cams. A few representative features such as mean and variance were selected and transformed to the relevant form for the classifier. Classification experiments were performed with the developed classifier and data gathered from the test bed. According to assigned K values, a successful classification result was obtained which means appropriate failure recognition.

Cyber-Physical Production Systems (CPPS), which pursue the implementation of machine intelligence in manufacturing systems, receive much attention as an advanced technology in Smart Factories. CPPS significantly necessitates the selflearning capability because this capability enables manufacturing objects to foresee performance results during their process planning activities and thus to make data-driven autonomous and collaborative decisions. The present work designs and implements a self-learning factory mechanism, which performs predictive process planning for energy reduction in metal cutting industries based on a hybrid-learning approach. The hybrid-learning approach is designed to accommodate traditional machine-learning and transfer-learning, thereby providing the ability of predictive modeling in both data sufficient and insufficient environments. Those manufacturing objects are agentized under the paradigm of Holonic Manufacturing Systems to determine the best energy-efficient machine tool through their self-decisions and interactions without the intervention of humans’ decisions. For such purpose, this paper includes: the proposition of the hybrid-learning approach, the design of system architecture and operational procedure for the self-learning factory, and the implementation of a prototype system.