This study aims to diagnose the early-stage degradation of 18650 lithium-ion batteries by conducting 20 constant-current/constant-voltage (CC-CV) charge-discharge cycles on both new (N.C.) and used (U.C.) cells. During the charge-discharge processes, strain was measured by attaching a strain gauge to the central region of the cell casing, while the voltage response was simultaneously monitored. The N.C. exhibited nearly constant strain behavior and reversible elastic deformation throughout all cycles. In contrast, the U.C. showed an increase in maximum strain, accumulation of residual strain, and pronounced nonlinear hysteresis, confirming the progression of internal structural degradation. Although both cells maintained similar voltage ranges under protection circuit module (PCM) control and appeared to show no apparent performance degradation, the strain response of the U.C. clearly revealed early degradation signals. These findings suggest that casing strain-based measurements provide an effective and non-destructive approach for diagnosing internal damage in lithium-ion batteries that is difficult to identify using electrical signals alone.
Prior-art search in patent examination is challenging due to the mismatch between IPC (text-based) and Locarno (visual/functional) taxonomies. This study establishes a corpus by matching Korean patent/ utility-model documents with Locarno codes. We propose a multimodal classifier that integrates IPC, text, and images. The text branch combines KorPatBERT embeddings with embeddings of the IPC hierarchy (section/class/subclass). The image branch fuses AlexNet features with local binary pattern and adaptive hierarchical density histogram descriptors. By comparing text+IPC, image+IPC, and text+image+IPC settings, we demonstrate that the fusion model consistently outperforms unimodal baselines. IPC injection and visual cues improve discrimination for sparse or noisy claims and for visually similar Locarno classes.
This study presents a fault diagnosis method utilizing deep learning-based vibration signal analysis to ensure the safety of electric vehicles equipped with in-wheel motors (IWMs). IWMs are susceptible to road-induced impacts, which can degrade power performance and compromise vehicle safety. Consequently, three-axis accelerometers were installed on the vehicle's upper arm, brake caliper, and mounting bolts to collect vibration data under normal-road and speed-bump driving conditions. The normal condition was defined as stable operation without abnormalities, while the fault condition included electrical and mechanical faults such as inverter damage and W-phase short circuits. The acquired vibration signals were processed using a bandpass filter and converted into log-Mel spectrograms, which were then classified using a Convolutional Neural Network (CNN) model with transfer learning. Experimental results from real vehicle tests demonstrated an average classification accuracy of 98.80% for four driving condition classes: normal-road, normal-bump, fault-road, and fault-bump. These results confirm the applicability of deep learning-based vibration analysis for IWM fault diagnosis.
In this paper, we propose an inverse method using proper orthogonal decomposition (POD) to obtain the boundary loads of elastic bodies from measured displacements. Reduced POD vectors are constructed from snapshot displacements for various training loads, and an inverse formulation is developed to estimate the boundary loads from measured displacements. Numerical results show that the error in the estimated loads decreases as the measurement location approaches the loading location. In addition, the error tends to steadily decrease when the number of POD modes increases. The proposed inverse method can stably estimate boundary loads from measured displacements, and can be used to estimate unknown loads and evaluate residual stresses in structures.
Automatic balancing control is a technique to reduce synchronous control current due to mass unbalance using active magnetic bearings (AMBs). It uses a synchronous notch filter in addition to a levitation control, with the filter frequency equal to the rotor speed. An alternative to a separate speed sensor is to use a speed estimator utilizing the synchronous component in vibration signals. To overcome the complexity of previous efforts on speed estimation where two orthogonal vibration signals are needed, we proposed a simple estimator based on phase-locked loop (PLL) applied to a single vibration signal. The loop filter in PLL is designed to obtain the target performance of the estimation. The effects of spurious signals having the frequency twice of the input signal is also investigated. The speed estimator is implemented on a test rig, demonstrating an estimation accuracy of 96.5%. The synchronous notch filter using the estimated speed is added to the levitation control, resulting in more than 90% reduction in synchronous control currents.
