Industry 4.0 is transforming predictive failure management by utilizing deep learning to enhance maintenance strategies and automate production processes. Traditional methods often fail to predict failures in time. This research addresses this issue by developing a time-to-fault prediction framework that utilizes an enhanced long short-term memory (LSTM) model to predict machine faults. The proposed method integrates real-time sensor data, including current, voltage, and temperature calibrated via ultra-sensitive optical sensing technologies based on the typical whispering gallery optical mode (WGM) to create a robust dataset. Due to the high-quality factor that these sensors exhibit, any minute change on the surrounding medium will makes a significant change on its transmission spectrum. The LSTM model trained on these data demonstrated rapid and stable convergence, outperforming other deep learning techniques with a mean absolute error (MAE) of 0.83, a root mean squared error (RMSE) of 1.62, and a coefficient of determination (R²) of 0.99. The results show the superior performance of LSTM in predicting machine failures early in real-world environments within 10 min lead time, improving productivity and reducing downtime. This framework advances smart industries by improving fault prediction in manufacturing precision robotics components, demonstrated through two humanoid robots, GUCnoid 1.0 and ARAtronica.

In this paper, an optical vibrometer/accelerometer is designed using micro-optical dielectric sensors based on whispering gallery modes (WGM) phenomena. The proposed design is created as a sphere made from polydimethylsiloxane (PDMS). The dielectric sensor would be laid between a piezo stack and a proof mass of ${10}^{-4}\mathrm{kg}$ (implemented as body force to the sensor) in the vertical position. Once the piezo stack starts to impose a vibration in the vertical direction, the proof mass starts to vibrate, leading to a change in the morphology of the sensor. In turn, the WGM shifts would be seen on the transmission spectrum and would be related to the vibration/acceleration of the piezo stack. An analysis based on the force transmissibility from the piezo stack to the optical sensor and calibrations are carried out along with preliminary designs and experiments. Results showed that the proposed sensor could reach a very high resolution up to 8 nano-g with sensitivity $d\lambda /da\approx 2.33\mathrm{pm}/\mathrm{\mu g}$.

Fault diagnosis in industrial systems remains a critical challenge due to complex operational conditions and the need for high accuracy in real-time applications. Traditional deep learning models often struggle with generalization and adaptability, leading to misclassifications, reduced accuracy, and unreliable fault detection. To overcome these limitations, this study proposes an enhanced hybrid Transformer-deep neural network (DNN) architecture designed for both binary and multi-class fault classification in industrial environments. By leveraging the Transformer’s superior feature extraction capabilities alongside the DNN’s classification strength, the model significantly improves detection accuracy and reduces misclassification errors. The proposed model is evaluated on two distinct datasets, including industrial machine fault detection (TMFD) and induction motor fault diagnosis (MFD). In binary classification tasks, the hybrid Transformer-DNN achieved accuracies of 99.97% for industrial machines and 98.26% for induction motors. For multi-class classification, it attained 99.97% accuracy on industrial machines and 98.39% on induction motors. These results demonstrate the model’s superior performance and robust fault identification across diverse datasets compared to other models. Furthermore, the system’s outputs are integrated into a humanoid-based telepresence platform named ARAtronica, facilitating enhanced remote monitoring and safety during fault inspections. This research offers a scalable, highly accurate framework to advance predictive maintenance in various industrial applications.

This paper investigates the use of artificial neural networks (ANNs) as a viable digital twin or alternative to the typical whispering gallery mode (WGM) optical sensors in engineering systems, especially in dynamic environments like robotics. Because of its fragility and limited endurance, the WGM sensor which is based on micro-optical resonators is inappropriate in these kinds of situations. In order to address these issues, the paper suggests an ANN that is specifically designed for the system and makes use of the WGM sensor’s high-quality factor ( Q -factor). By extending the applicability and endurance to dynamic contexts and reducing fragility problems, the ANN seeks to give high-resolution measurement. In order to minimize post-processing requirements and maintain system robustness, the study goal is for the ANN to function as a representative predictor of the WGM sensor output. The GUCnoid 1.0 humanoid robot is used in the paper as an example to show how the WGM optical sensors may improve humanoid robot performance for a variety of applications. The results of the experiments demonstrate that the sensitivity, precision, and resolution of ANN outputs and actual WGM shifts are equivalent. As a consequence, current obstacles to the widespread use of high-precision sensing in the robotics industry are removed, and the potential of ANNs as virtual substitutes or the digital twin for genuine WGM sensors in robotics systems is validated. So, this paper can be very beneficial not only to the sensing technologies that are used in robotics, which are subjected to the dynamic environments, but also to the industrial automation and human-machine interface.

In this paper, a novel technique using an ultra-sensitive optical resonator based on whispering gallery modes (WGM) is proposed to detect the diffusion of organic solvents. The sensor configuration is a micro-cavity made of polymeric material. When the solvent starts to diffuse, the polymer of the cavity starts to swallow that solvent. A swollen elastomer is in fact a solution, except that its mechanical response is now elastic rather than viscous. As solvents fill the network, chains are extended. In turn, that leads to the change of the morphology and mechanical properties of the sensing element. These changes could be measured by tracking the WGM shifts. Several experiments were carried out to measure that swelling force. Ethanol and methanol are used in this paper as candidates to study their driving force of diffusion (concentration gradient) on the cavity. Additionally, this sensing design can be used for biological sensing application. Breath diagnosis can use this configuration in diabetes diagnosis since a solvent like acetone concentration in human breath leads to a quick, convenient, accurate, and painless breath diagnosis of diabetes. The optical resonator results are verified through two different analyses: theoretical and experimental modeling. These micro-optical cavities have been examined using preliminary experiments to fully investigate their response and to verify the numerical analysis. Results show that the proposed sensor yields sensitivity for the driving force of diffusion (concentration gradient) (9.405 × 10¹³ pm/N) with a measurement precision of ~3.6 fN.

Acoustic measurements are the obvious target of acoustic investigations in this paper, using an opto-mechanical frequency analyzer. This proposed sensor is designed using micro-optical dielectric sensors based on whispering gallery modes (WGMs). This phenomenon, commonly referred to as WGM, is excited by evanescently coupling light from a tunable laser diode using a tapered single-mode optical fiber. A measurement arrangement can typically be separated into source and receiver components. The receiver component consists of a sound level meter or sound analyzer that we call an opto-mechanical frequency analyzer, which displays the total sound level in frequency-dependent data. It can also produce results such as spectra or impulse responses. Furthermore, measurements are based on special measurement environments that allow several reference sound fields to be created. The proposed design is composed of a tuner that consists of slots with different geometries with dielectric beams placed on each one. These beams are made from polydimethylsiloxane. The spherical optical polymeric resonators are mechanically coupled to the dielectric beams placed at each slot. The tuner has a different resonant frequency at each slot, which depends on the effective length of the beam. Sound waves deform the polymeric cavity, causing a shift in its transmission spectrum. A technique based on signal cross-correlation is used to calculate that shift, which is known as WGM shift. An analysis and calibration are carried out along with preliminary designs and experiments. Results prove that the proposed technique can be used as a very high-resolution frequency analyzer due to the high quality factor ($Q$-factor) of the resonators, which is ${10}^6$ compared to the typical electrical frequency analyzer that has a $Q$-factor up to ${10}^2$. Similar to an optical prism, this device can be used to split sound into its constituent frequencies.