This study introduces a non-invasive approach to monitor operation and productivity of a legacy pipe bending machine in real-time based on a lightweight convolutional neural network (CNN) model and internal sound as input data. Various sensors were deployed to determine the optimal sensor type and placement, and labels for training and testing the CNN model were generated through the meticulous collection of sound data in conjunction with webcam videos. The CNN model, which was optimized through hyperparameter tuning via grid search and utilized feature extraction using Log-Mel spectrogram, demonstrated notable prediction accuracies in the test. However, when applied in a real-world manufacturing scenario, the model encountered a significant number of errors in predicting productivity. To navigate through this challenge and enhance the predictive accuracy of the system, a buffer algorithm using the inferences of CNN models was proposed. This algorithm employs a queuing method for continuous sound monitoring securing robust predictions, refines the interpretation of the CNN model inferences, and enhances prediction outcomes in actual implementation where accuracy of monitoring productivity information is crucial. The proposed lightweight CNN model alongside the buffer algorithm was successfully deployed on an edge computer, enabling real-time remote monitoring.

In this research, we propose an Error Separation Technique (EST) based on optical fiber sensors for on-machine straightness error measurement. Two fiber optic Fabry–Pérot interferometers have been developed serving as two displacement sensors. The displacement distance is computed according to the reflected spectrum from interferometers, which can achieve a sub-micrometer resolution. The two-point method has been employed to separate the straightness error of the slides and the profile error of a fine-polished standard block. The spacing distance between two interferometers is determined by the diameter of optical fibers so that the EST's resolution has the potential to reach the sub-millimeter scale. In the experiment, the straightness error has been measured on a commercially available computer numerical control machine tool, and the measurement has been conducted on its x-axis. The spacing distance between two optical fiber sensors is 1.5 mm which equals the EST's resolution along the machine tool's x-axis. The separated profile error of the measured standard block is around 30 µm which has been verified by a high precision Coordinate Measurement Machine (CMM). The magnitude of the separated straightness error is around 40 µm. This technique is flexible and simple to be conducted, which can contribute to the micro-machine tool calibration and other straightness error applications.

With the new era of the fourth industrial revolution, adopting human skills and knowledge is getting more valuable to enable smart manufacturing. Moreover, a cognitive ability that comes from human sensory and intuition is able to facilitate smart sensing and collaboration between humans and autonomy. For smart manufacturing, the improvement of collaboration of humans, autonomy, and physical systems is essential. In this paper, an immersive and interactive cyber-physical system (I ² CPS) is suggested to achieve efficient collaboration between humans and autonomy. Virtual reality (VR) with a game engine, a new middleware, and a data-to-information protocol were introduced for I ² CPS. This system was utilized in an autonomous robot grasping platform constructed by human skills and self-learning. The initial convolutional neural network (CNN) model was trained by manual demonstration and showed 39% of the success rate of grasping blocks. After self-learning, the success rate increased to 70%. As another human-inspired sensing technique, sound recognition system using ambient sound and the CNN model to predict the running state of the generator is proposed. The Mel spectrum based on human hearings was used as feature extraction. The overall running prediction accuracy of the evaluation dataset showed approximately 95%. This system was applied for web-based remote and real-time monitoring of generators.

This paper presents the Multi-Sensor and MTConnect (MSM) dataset, an open-access resource for anomaly detection in computer numerical control (CNC) metal milling. The dataset integrates synchronized signals from sound sensors, accelerometers, current transformers, and MTConnect-based machine controller data, collected from both laboratory experiments and real industrial production. It covers diverse machining conditions, including normal operations, process anomalies, and tool defects. All data were reviewed and annotated by domain experts using a three-level scheme, enabling consistent labeling across machines and environments. The dataset spans multiple CNC mills, cutting tools, workpiece materials, and cutting conditions, and includes MTConnect information models to ensure semantic consistency and reproducibility. Each dataset unit is time-aligned across sensor modalities, allowing direct use in multimodal analysis. By combining heterogeneous sensors with standardized machine data, the MSM dataset provides a reusable benchmark for artificial intelligence (AI)-based monitoring, anomaly detection, and research in smart manufacturing.