This study presents an integrated approach that combines optical sensing and deep learning for reliable defect detection in eco-friendly molded pulp packaging. The You Only Look Once (YOLO) version 8 deep learning model was applied to automate post-production inspection and improve the consistency of manual checks. The process involved image acquisition, annotation, dataset partitioning, model training, and simulated edge deployment. The proposed model achieved a validation accuracy of 98.75%, demonstrating a strong ability to identify surface defects. This result confirms the effectiveness of optical sensing–based deep learning for real-time and precise inspection in sustainable packaging manufacturing.

This study presents an AI-assisted long-distance fiber Bragg grating (FBG)-based sensing approach for simultaneous temperature and vibration measurement using a single bare FBG sensor. To address the strong coupling between temperature- and vibration-induced effects in the wavelength time series, a signal processing framework based on adaptive variational mode decomposition (AVMD) is developed. With power-spectral-density-guided parameter selection, the mixed wavelength signal is separated into a low-frequency temperature-related component and a high-frequency vibration-related component, enabling stable temperature-vibration decoupling within a single-sensor architecture. Experiments conducted with a 10 km fiber link between the sensor and the interrogator demonstrate that the proposed method can stably track the dominant vibration frequency under various temperature and vibration conditions, while the reconstructed low-frequency component remains consistent with the thermal evolution trend even in the presence of vibration. Random vibration tests and low-frequency vibration resolution analysis further confirm the stability and practicality of the proposed approach under long-distance fiber transmission conditions. In addition, an AI-assisted condition-monitoring scheme is demonstrated using a one-dimensional convolutional autoencoder trained solely with normal wavelength time-series data. Rather than relying on raw reconstruction error alone, the diagnostic layer derives a latent transition score from encoder bottleneck features through temporal pooling, L2 normalization, cosine-distance evaluation, smoothing, and baseline removal. Deviations from steady operating conditions can thereby be preliminarily indicated, highlighting the potential for integrating physics-driven signal processing with data-driven artificial intelligence in long-distance fiber sensing systems.

Free-space optical (FSO) communication offers high bandwidth, license-free spectrum, and fast deployment, but is highly sensitive to beam misalignment caused by mechanical vibrations and external disturbances, which can severely reduce the received optical power. To improve robustness and energy efficiency, this study presents a reinforcement learning (RL)-based automatic beam alignment control framework for FSO systems. Real-time measurements of the beam spot position and received optical power are used as the system state, while the RL controller generates continuous beam-steering actions. Three continuous-action algorithms, namely deep deterministic policy gradient (DDPG), twin delayed DDPG (TD3), and soft actor-critic (SAC), are compared in terms of convergence, robustness, and control energy. Simulation results show that SAC achieves the lowest control energy while maintaining the fastest average recovery among the three algorithms.

This paper introduces ensemble deep learning (EDL) models for accurate demodulation of Fiber Bragg Grating (FBG) sensor spectra, with input training data enhanced through the numerical square method (NSM) technique. The method combines numerical squaring with EDL to improve prediction accuracy under noisy and overlapping spectral conditions. The proposed system uses an FBG interrogator with sensors embedded in transparent liquid containers to detect liquid level changes via Bragg wavelength shifts. However, changes in the liquid level shift the float position, causing overlap or cross-talk between adjacent sensors. To address this, the NSM enhances signal clarity by boosting authentic FBG responses and minimizing noise. Single, double, and triple NSM operations are applied to the experimental data to sharpen the reflection peaks and enhance overall signal quality, ensuring the data are well-prepared for training the EDL model. The EDL model combines CNNs for spectral feature extraction and LSTMs and GRUs for sequential pattern recognition and efficient learning, effectively capturing both spectral and sequential features. The EDL model is trained on the dataset enhanced by the NSM technique and validated using unseen experimental data to evaluate its performance. The results confirm that NSM efficiently sharpens reflection peaks, improves signal clarity, and reduces noise in FBG spectra, while the EDL model effectively demodulates the overlapping spectra. Experimental results demonstrate that the integrated NSM-EDL approach outperforms traditional machine learning and standalone deep learning models in terms of prediction accuracy, minimal errors, and computational time. The proposed method is cost-effective, hardware-efficient, and ideal for real-time multiplexed FBG sensing applications, such as liquid level monitoring.

This paper proposes a large-scale, self-healing multipoint fiber Bragg grating (FBG) sensor network that employs reinforcement learning (RL) techniques to enhance the resilience and efficiency of optical wireless communication networks. The system features a mesh-structured, self-healing ring-mesh architecture employing 2 x 2 optical switches, enabling robust multipoint sensing and fault tolerance in the event of one or more link failures. To further extend network coverage and support distributed deployment scenarios, free-space optical (FSO) links are integrated as wireless optical backhaul between central offices and remote monitoring sites, including structural health, renewable energy, and transportation systems. These FSO links offer high-speed, line-of-sight connections that complement physical fiber infrastructure, particularly in locations where cable deployment is impractical. Additionally, RL-based artificial intelligence (AI) techniques are employed to enable intelligent path selection, optimize routing, and enhance network reliability. Experimental results confirm that the RL-based approach effectively identifies optimal sensing paths among multiple routing options, both wired and wireless, resulting in reduced energy consumption, extended sensor network lifespan, and improved transmission delay. The proposed hybrid FSO-fiber self-healing sensor system demonstrates high survivability, scalability, and low routing path loss, making it a strong candidate for future services and mission-critical applications.

This study demonstrates single-channel fiber Bragg grating (FBG) sensing for relative vibration-state monitoring of a motor-support system under angle-dependent boundary conditions. A packaged FBG accelerometer-type sensing unit was mounted on the motor-support structure, and the reflected Bragg wavelength was recorded as a one-dimensional optical vibration response. Because the sensor was installed away from the rotating shaft, the measured wavelength fluctuation was interpreted as a coupled vibration-sensitive response of the motor, fixture, sensor package, bonding condition, and changing boundary state, rather than as a calibrated shaft speed or absolute acceleration signal. Adaptive variational mode decomposition (AVMD) was applied to track the time-varying narrowband spectral-response trajectory of the Bragg-wavelength signal. In parallel, raw wavelength windows were supplied to LSTM, 1D-CNN, and CNN-LSTM autoencoders to evaluate waveform departures from learned nominal fixed-angle behavior. The fixed-angle results showed stable but distinguishable optical vibration responses under different boundary states, whereas the dynamic angle-transition records produced local trajectory changes and alarm-candidate intervals. Baseline and autoencoder comparisons further clarified the trade-off between transition coverage and false-alarm tendency. The RMS threshold baseline was more sensitive to transition-related amplitude changes but produced more false alarms, whereas the CNN-LSTM autoencoder provided the most selective response among the tested autoencoder branches. The results are interpreted as task-specific evidence for relative vibration-state transition monitoring rather than as general motor fault diagnosis. Overall, the framework demonstrates a compact FBG-based route for relative vibration-state transition monitoring when speed references, dense sensor layouts, and labeled fault data are unavailable.

Fifth- and sixth-generation (5G/6G) mobile networks demand advanced beamforming for millimeter-wave (mmWave) massive MIMO systems. Conventional electronic beamforming suffers from high complexity, power consumption, and latency. This paper presents an inverse design framework using machine learning (ML) for photonic beamforming networks, enabling intelligent control of variable optical delay lines (VDLs) integrated with array antennas. The method optimizes beam directivity for single-user and enables multi-user (MU-MIMO) beamforming. Furthermore, deep neural networks (DNN), tabular data neural network (TabNet), traditional neural networks (NN), long short term memory networks (LSTM), extreme gradient boosting (XGBoost), and lightweight gradient boosting machine (LightGBM) are involved in the proposed inverse design ML models for thorough comparison. Results show LightGBM achieves the highest accuracy for VDL prediction due to its tabular data processing efficiency. This work provides a foundation for low-complexity, low-power photonic front-ends for 5G/6G.

This article presents an advanced sensor data processing framework leveraging a hybrid deep learning network (DLN) composed of multilayer perceptron (MLP) and convolutional neural network (CNN) models to accurately detect, classify, and reconstruct overlapping temperature events in distributed temperature sensing (DTS) systems. DTS systems frequently face challenges related to limited spatial resolution and overlapping thermal profiles, significantly impairing accurate event detection and localization in different applications. To overcome these limitations, we propose a novel sensor data fusion and pattern recognition approach employing simulated and experimental DTS datasets. Our hybrid DLN extracts intricate features from sensor data, effectively reconstructing temperature profiles with minimal gaps of 0.1 m between events, achieving a mean absolute error (MAE) of 0.104 m. The proposed method demonstrates robust generalization capabilities and high accuracy in real-world industry application scenarios, significantly enhancing the sensor's data processing capability without necessitating modifications to existing DTS infrastructure. This research provides substantial advancements in soft computing methodologies for sensor data processing, particularly in high-density thermal event detection and classification.

This paper presents a Fiber Bragg grating (FBG) sensor-integrated robotic-arm sensing framework for simultaneous motion classification and joint-angle regression using heterogeneous stacked ensemble generalization learning (HSEGL). Two embedded FBG sensors capture strain-induced Bragg wavelength shifts from a three-segment robotic arm during bidirectional joint movements. Extracting reliable motion information from FBG spectra remains challenging because of spectral overlap, sensor cross-talk, vibration-induced wavelength fluctuations, nonlinear strain responses, and limited experimental data. To address these issues, spectral responses from the two sensors are fused across three motion scenarios, preprocessed, and normalized. The resulting data are then provided as inputs to the HSEGL base learners: a one-dimensional convolutional neural network (1D CNN) for local spectral feature extraction, a gated recurrent unit (GRU) for sequential dependency learning, and a deep neural network (DNN) for global nonlinear feature representation. Their predictions are combined through mean and weighted ensemble strategies and concatenated as meta-features for stacked generalization. Model robustness is evaluated using repeated stratified five-fold cross-validation with multiple random seeds. HSEGL achieves the highest average validation accuracy, precision, recall, and F1-score, together with the lowest performance variation and joint-angle regression errors among the evaluated models. These findings demonstrate accurate and consistent robotic-arm sensing under limited data and controlled laboratory conditions.

Conceptually, this paper aims to help reduce the communication blind spots originating from the design of millimeter-wave (mmW) beamforming by deploying radio units of an open radio access network (O-RAN) with free-space optics (FSOs) as the backhaul and the fiber-optic link as the fronthaul. At frequencies exceeding 24 GHz, the transmission reach of 5G/6G beamforming is limited to a few hundred meters, and the periphery area of the sector operational range of beamforming introduces a communication blind spot. Using FSOs as the backhaul and a fiber-optic link as the fronthaul, O-RAN empowers the radio unit to extend over greater distances to supplement the communication range that mmW beamforming cannot adequately cover. Notably, O-RAN is a prime example of next-generation wireless networks renowned for their adaptability and open architecture to enhance the cost-effectiveness of this integration. A 200 meter-long FSO link for backhaul and a fiber-optic link of up to 10 km for fronthaul were erected, thereby enabling the reach of communication services from urban centers to suburban and remote rural areas. Furthermore, in the context of beamforming, reinforcement learning (RL) was employed to optimize the error vector magnitude (EVM) by dynamically adjusting the beamforming phase based on the communication user’s location. In summary, the integration of RL-based mmW beamforming with the proposed O-RAN communication setup is operational. It lends scalability and cost-effectiveness to current and future communication infrastructures in urban, peri-urban, and rural areas.

In the context of significant climate change, monitoring inclination, water levels, and temperatures in public buildings and surrounding environments is sensible. This paper presents a pair of fiber Bragg grating (FBG) subsidence sensor systems designed to simultaneously measure tilt and water levels and explore the system’s potential to detect temperature variations. The configuration of the FBG subsidence sensor is intentionally skewed to enhance measurement sensitivity. The system is capable of concurrently detecting a 0.5 cm variation in water level and a 0.424° change in tilt, with tilt measurements spanning from −1.696° to 1.696°. Furthermore, the measurement system can be integrated with free-space optics (FSO), which is anticipated to address the challenges associated with installing fiber optic cables. Consequently, the proposed innovative FBG sensor system can measure multiple parameters using fewer sensors, thereby improving sensing capacity and cost-efficiency.

In the modern world, robots have become increasingly essential across various industries. Activity monitoring has emerged as a key method for diagnosing environmental conditions and enhancing the intelligence of mechanical robots. Vibration or strain from different activities is a critical parameter for evaluating and detecting activities, which presents a significant challenge in accurately assessing robotic performance across diverse tasks. This article demonstrates a novel method for activity monitoring of mechanical robot-dog machines that prevents motor wear, reduces high maintenance costs, and increases the durability of machines. The method utilizes an optical fiber-based fiber Bragg grating (FBG) sensor system to detect dynamic strain resulting from vibration signals generated by robotic movements, ensuring precise monitoring of robotic dog activities and the you only look once version 7 (YOLO-v7) algorithm for activity detection. Two model modules are implemented: the first experiment collects dynamic strain data for up to eight possible activities, and the second experiment collects the five different weights dragged by the robot dog. YOLO-v7 ensures and evaluates robot activities. The detection results demonstrate the eight activities and different weight-carrying model accuracy of 95.31% and 98.81%, respectively. The model performance shows that strains from the motor machine are accurately detected, signaling anomalies. Thus, our proposed experimental setup is flexible, cost-effective, robust, computationally efficient, fast, and improves the sensing quality of robot pose action monitoring

We present a novel fiber Bragg grating (FBG) sensor system that employs both reflective and transmission wavelengths to significantly boost capacity within a single network. By integrating free-space optics (FSOs) for remote communication, this architecture operates beyond physical constraints where laying fiber is not feasible, thereby enhancing deployment flexibility. To address the overlap of Bragg wavelengths commonly encountered in dense FBG networks, we utilize a Transformer convolutional neural network (CNN) that leverages positional encoding to more effectively learn data patterns. Preliminary results demonstrate a significant enhancement in sensor capacity and detection accuracy. Moreover, the proposed Transformer CNN reduces both training and inference times by approximately 2.24x and 1.43x, respectively, compared to traditional Transformer models. By leveraging both reflective and transmission FBG readouts within a fixed spectral window, the system supports nearly double the number of sensors, offering a scalable and efficient solution well-suited for high-density, real-time industrial sensing applications.

Making the best use of expensive distributed temperature sensing (DTS) systems and saving costs is helpful for their measurement applications. The integration of free space optics (FSO) and DTS systems helps to reduce the cost of laying fiber optic cables and maintaining optical cables. In addition, machine learning is a common means to optimize the measurement demodulation of distributed fiber sensing. This work uses meta-learning to demodulate the measurement results of DTS. Meta-learning allows the use of small amounts of data to work, thereby reducing the data modeling cost and time of actual measurement applications. The atmospheric turbulence caused by the introduction of FSO or the attenuation or scattering caused by long-distance optical fibers may introduce noise into the temperature distribution curve measured by DTS. Therefore, this work also simulates the temperature distribution curve with noise and applies the discrete wavelet transform (DWT) to remove noise to reduce the data quality degradation caused by noise. Results show that the proposed meta-learning model can adapt faster and significantly improve the temperature event detection performance. Therefore, the proposed FSO and meta-learning system are cost-effective for applications in DTS and can enhance the distributed fiber optic sensing system.

In this article, we propose a novel method that integrates deep learning with Fabry-Perot liquid crystal (FP-LC) technology for fiber Bragg grating (FBG) interrogation. The use of FP-LC enhances the measurement range and enables high sensitivity in FBG sensors, making them appropriate for a wide range of applications requiring precise and responsive sensing. However, collecting a large amount of real experimental FBG sensor data is time-consuming, technically challenging, and resource-intensive. To address this issue, we utilize a conditional generative adversarial network (CGAN) to generate a sufficient amount of synthetic training data. The CGAN generates data conditioned on real FBG sensor data, ensuring that the generated data closely look like real experimental data distributions, which is crucial for effective model training. Moreover, we proposed a convolutional neural network (CNN) method to solve crosstalk problems, to improve sensing accuracy, and to precisely detect the peak wavelength of each FBG sensor. The experimental results demonstrated that the proposed CGAN technique effectively generates a large amount of data to improve the performance of the proposed CNN model. Furthermore, the results proved that the CNN trained on CGAN-generated data significantly improves the detection speed and accuracy of central wavelength measurements compared to traditional approaches. Hence, the proposed system is cost-effective, easy to set up for experiments, increases the feasibility and portability of modularization, fast and flexible, overcoming data shortages, and improving the sensing accuracy of wavelength detection for FBG sensor systems.

A combination of optical Fiber Bragg Gratings (FBGs) and a Tunable Delay Line Interferometer (TDLI) is proposed for capturing vibration signals amidst inherent, undeniable noises. The proposed system leverages the adjustable Free Spectral Range (FSR) of the TDLI to significantly enhance the sensitivity of FBG-based sensors, particularly for distinguishing extremely close superposition vibration frequencies in various frequency ranges. The tuning technique for FSR adjustment in FBG peak power modulation, designed to achieve accurate measurement signals with diminished noise when FBG wavelength shift occurs, is revealed. Incorporating Free-Space Optics (FSOs) into the vibration sensing framework further enhances system flexibility by facilitating wireless, high-speed vibration data transmission in scenarios where direct fiber connections are impractical, such as remote or dynamic environments. Moreover, the system embeds deep learning algorithms that enable high-accuracy recognition of vibration frequencies without the need for denoising techniques. The coalescences of FBG-TDLI and FSO technologies, integrated with deep learning techniques, create a highly precise, adaptable, and efficient vibration sensing system. The findings underscore this approach’s potential to redefine vibration sensing applications in critical infrastructure, power plants, and industrial systems, offering a new paradigm for precision monitoring in challenging operational conditions.

This study introduces a novel deep neural network (DNN) framework tailored to breaking the sampling limit for high-frequency vibration recognition using fiber Bragg grating (FBG) sensors in conjunction with low-power, low-sampling-rate FBG interrogators. These interrogators, while energy-efficient, are inherently limited by constrained acquisition rates, leading to severe undersampling and the obfuscation of fine spectral details essential for accurate vibration analysis. The proposed method circumvents this limitation by operating solely on raw time-domain signals, learning to recognize high-frequency and extremely close vibrational components accurately. Extensive validation using the combination of simulated and experimental datasets demonstrates the model’s superiority in frequency discrimination across a broad vibrational spectrum. This approach is expected to be a significant advancement in intelligent optical vibration sensing and compact, low-power condition monitoring solutions in complex environments.

This paper introduces a novel Fiber Bragg Grating (FBG) based optical fiber sensing system integrated with deep learning (DL) for an enhanced liquid-level sensing system. The proposed system utilized FBG sensors to improve liquid level measurement sensitivity, accuracy, and cost-effectiveness. The proposed FBG sensor-based liquid level sensing system monitors liquid levels by detecting Bragg wavelength shifts, which occur due to changes in the liquid level inside the containers, as controlled by the system. However, in sensor multiplexing, changing the liquid level inside the container can lead to reflected signal overlap or cross-talk between the sensors, potentially affecting the sensitivity and accuracy of the measurements. A strong linear relationship has been observed between strain sensitivity and the liquid level in this proposed FBG-based liquid-level sensing system. To address the issues of the reflected signal overlap between sensors and accurately predict the liquid level in each container, this study proposes an ensemble deep learning (EDL) approach. The performance of the well-trained EDL model demonstrates that it can accurately predict liquid levels at each point. Experimental results prove that the model achieves excellent detection accuracy, extremely low testing times, and minimal error values, and significantly improves the overall accuracy of the liquid-level sensing system. Moreover, the proposed EDL model outperforms other DL models in prediction accuracy. The proposed EDL model offers a strong potential for creating an accurate, cost-effective liquid-level sensing system with high accuracy.

This study presents the liquid crystal Fabry–Pérot etalon (LC-FP) as the preferred laser wavelength tuning solution within a erbium-doped fiber ring laser architecture. The laser cavity wavelength can be adjusted by applying varying voltages to the LC-FP. Furthermore, tuning the laser wavelength can be facilitated by modifying the incident light through changes in the steering angle of the LC-FP, which is attributed to the angular dispersion characteristics of the device. The operational range for the steering angle of the LC-FP is ± 4 to 18 degrees. This architectural framework is adept at facilitating the generation of single-wavelength and dual-wavelength lasers within the C band. The tunable range for a single wavelength is approximately 13 nm, while the tunable range for dual wavelengths is around 14 nm, with a wavelength spacing of approximately 17.5 nm. These capabilities are primarily influenced by the operational wavelength of the erbium-doped fiber amplifier (EDFA), the operating wavelength of the collimator that directs the fiber optic beam into the LC-FP, and the fixed thickness of the LC-FP.

Fiber Bragg grating (FBG) sensing systems face significant challenges in resolving overlapping spectral signatures when multiple sensors operate within limited wavelength ranges, severely limiting sensor density and network scalability. This study introduces a novel Transformer-based neural network architecture that effectively resolves spectral overlap in both uniform and mixed-linewidth FBG sensor arrays, operating under bidirectional drift. The system uniquely combines dual-linewidth configurations with reflection and transmission mode fusion to enhance demodulation accuracy and sensing capacity. By integrating cloud computing, the model enables scalable deployment and near-real-time inference even in large-scale monitoring environments. The proposed approach supports self-healing functionality through dynamic switching between spectral modes during fiber breaks and enhances resilience against spectral congestion. Comprehensive evaluation across twelve drift scenarios demonstrates exceptional demodulation performance under severe spectral overlap conditions that challenge conventional peak-finding algorithms. This breakthrough establishes a new paradigm for high-density, distributed FBG sensing networks applicable to land monitoring, soil stability assessment, groundwater detection, maritime surveillance, and smart agriculture.