QTQin Tang

Qin Tang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Qin Tang has a total of 5 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Label-Free Physics-Guided Transfer Learning for Cross-Scale Lithium-Ion Battery SOH Estimation

    by Getachew Shimelash Abebe, Bin Gou, Qiujing Peng, Qin Tang

    Accurate state-of-health estimation for lithium-ion battery packs remains challenging because real-world packs rarely provide complete charge-discharge cycles, target-domain labels are costly to obtain, and laboratory cell data differ substantially from on-road pack behavior. Existing transfer-learning methods reduce labeling effort, but most are limited to cell-level validation, complete charging profiles, or source-to-target transfer settings that must be redesigned for each target battery. This study proposes a label-free cross-scale transfer-learning framework using partial charging data for practical SOH estimation while reducing degradation-testing cost. First, Physics-Informed Heterogeneous Source Unification (PI-HSU) constructs a unified source domain from two distinct public datasets using features with consistent physical meaning and harmonized labels. The framework then integrates domain-adversarial neural networks (DANN) with maximum mean discrepancy (MMD) to align the source and target domains, while Physics-guided test-time training (Battery TTT) enforces monotonicity, smoothness, boundary, and reference constraints. Finally, Failure-Aware Physics-Guided Pseudo-Label Ensemble (FAPPE) detects collapsed trajectories and reconstructs degradation from observable target features. FAPPE then stabilizes early- and late-life predictions by combining complementary pseudo-label predictors. Validated on six diverse datasets encompassing 36 batteries from different manufacturers, the framework achieved an average mean absolute error (MAE) of 2.69%. It was further tested on 20 on-road EV packs, achieving an average MAE of 1.16%, a 93.27% coverage rate, and an average interval width of 6.84. In addition, it was deployed on a Jetson edge-computing device, demonstrating its potential for industrial application. Deterministic and probabilistic validations confirm the practical applicability of the proposed framework without requiring additional target-domain degradation experiments.

  • Open Access

    Label-Free Physics-Guided Transfer Learning for Cross-Scale Lithium-Ion Battery SOH Estimation

    by Getachew Shimelash Abebe, Bin Gou, Qiujing Peng, Qin Tang

    Accurate state-of-health estimation for lithium-ion battery packs remains challenging because real-world packs rarely provide complete charge-discharge cycles, target-domain labels are costly to obtain, and laboratory cell data differ substantially from on-road pack behavior. Existing transfer-learning methods reduce labeling effort, but most are limited to cell-level validation, complete charging profiles, or source-to-target transfer settings that must be redesigned for each target battery. This study proposes a label-free cross-scale transfer-learning framework using partial charging data for practical SOH estimation while reducing degradation-testing cost. A unified source domain is constructed from two distinct public datasets by extracting physics-informed features from partial charging windows, avoiding the need for full charge-discharge cycles. The framework integrates domain-adversarial neural networks (DANN) with maximum mean discrepancy (MMD) to extract and align domain-invariant features. Physics-guided test-time training (Battery TTT) further adapts the model by enforcing monotonicity, smoothness, boundary, and reference constraints. In addition, exponential decay and extreme gradient boosting (XGBoost) ensemble models are combined to balance early- and late-cycle predictions in unlabeled target domains. Validated on six diverse datasets encompassing 36 batteries from different manufacturers, the framework achieved an average mean absolute error (MAE) of 2.69%. It was further tested on 20 real-world on-road EV packs, achieving an average MAE of 1.16%, a 93.27% coverage rate, and an average interval width of 6.84. In addition, it was successfully deployed on a Jetson edge-computing device, demonstrating its potential for industrial application. Deterministic and probabilistic validations confirm the practical applicability of the proposed framework without requiring additional target-domain degradation experiments.

  • Open Access

    Quadrotor Attitude Controller Design Based on Improved PID Control Algorithm

    by Qin Tang, Yibo Niu, Xiangkai Xu

    Quadrotor is a key technology widely used in national defence, electric power and agriculture. In order to solve the problems of response hysteresis and sensitivity to sudden input changes of the conventional Proportional-Integral-Derivative (PID) control algorithm in quadrotor control, a filter is added to the outer differential term based on the conventional serial PID control to form an incomplete differential algorithm, which improves the dynamic characteristics of the system and reduces the sensitivity to sudden signal changes; at the same time, a filter is added based on the conventional serial PID control algorithm to form an incomplete differential algorithm, which improves the dynamic characteristics of the system and reduces the sensitivity to sudden signal changes. A filter is added based on the conventional series PID control algorithm to form an incomplete differential algorithm, which improves the dynamic characteristics of the system and reduces the sensitivity to abrupt signal changes; at the same time, a filter is added based on the conventional series PID control algorithm to achieve a smooth controller response curve. Simulation results show that compared with the conventional PID controller, the improved PID control algorithm reduces the maximum overshoot of the system step response from the original 8% to 3%, and its robustness and reliability are greatly improved.