Abstract
The stochastic latencies induced by Controller Area Network (CAN) congestion, coupled with inherent actuator response lags, pose a serious threat to the active safety of networked electrified heavy-duty trucks (NEHTs), which function as complex cyber-physical systems relying on real-time data exchange. Unlike traditional methods, this paper presents a novel learning-based robust transient stability control framework resilient to coupled CAN and actuator delays (CCADs) by integrating risk warning, model learning, and delay-robust control to improve the vehicle driving stability and safety. Firstly, a precise and rapid-response warning approach based on transient load transfer index is proposed to accurately evaluate the transient dynamics of NEHTs and enable early instability warning, while effectively mitigating the impact of CCADs. Then, a data-driven time-delay model learning method is developed to capture the strong nonlinearity and operational uncertainties of NEHTs, and compensate for the residual dynamic mismatch that is often exacerbated under CCADs. Finally, a data-driven optimal vehicle control method against CCADs is proposed to improve the transient stability control effectiveness and robustness under various operating conditions. Hardware-in-the-loop experimental results demonstrate that the proposed method effectively ensures the transient stability of NEHTs, achieving the lowest total Integral of Absolute Error in every evaluated scenario, with reductions of up to 63.74% compared with conventional model predictive control, thereby enhancing the robustness under CCADs. Note to Practitioners - In modern NEHTs, stochastic delays from CAN congestion combined with inherent actuator response lags can severely degrade vehicle stability during emergency lane changes, high-speed cornering, or adverse conditions. These coupled delays often lead to dangerous jackknifing or rollover incidents, compromising the safety of both the truck and surrounding traffic. To solve this real-world problem, we developed a practical LRTSC framework. It integrates three easy-to-deploy modules: 1) a quadratic transient instability warning system based on load transfer dynamics that delivers earlier, more reliable alerts even when delays are present; 2) a data-driven Gaussian process learning module that uses onboard operational data to automatically correct model mismatches caused by network latencies; 3) a delay-aware optimal controller that coordinates active front-wheel steering, rear-axle torque vectoring, and trailer differential braking for smooth, stable response. Hardware-in-the-loop experiments on a 6× 4 semi-trailer truck show that the proposed method reduces the IAE by up to 63.74% compared with conventional model predictive control (MPC), while consistently outperforming it across all tested scenarios and maintaining stability under crosswinds and low-adhesion roads. The present validation is established on a hardware-in-the-loop platform with injected delays representative of typical CAN-bus and electric-actuator latency profiles, which confirms the effectiveness and real-time executability of the proposed framework under these conditions and demonstrates its promising potential for practical heavy-duty truck applications. Building on these results, truck manufacturers and control engineers can implement the framework on existing electronic control units to enhance active safety in next-generation connected electric heavy-duty trucks operating in real-world communication-constrained environments.
| Original language | English |
|---|---|
| Pages (from-to) | 16068-16084 |
| Number of pages | 17 |
| Journal | IEEE Transactions on Automation Science and Engineering |
| Volume | 23 |
| DOIs | |
| Publication status | Published - 2026 |
| Externally published | Yes |
Keywords
- Networked electrified heavy-duty trucks
- cyber-physical systems
- networked control systems
- robust control
Fingerprint
Dive into the research topics of 'Learning-Based Robust Transient Stability Control for Networked Electrified Heavy-Duty Trucks Under Coupled CAN and Actuator Delays'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver