在长途驾驶过程中，驾驶员腰部疲劳现象往往表现突出，成为影响汽车驾驶体验的一个主要方面，威胁驾驶员脊柱健康和驾驶安全。然而驾驶员腰部疲劳成因复杂，不易检测，无法为汽车座椅腰部舒适性优化设计提供支持。为解决这一问题，本文构建了基于驾驶员腰部肌肉内外特性的疲劳状态动态辨识算法，同时建立了基于肌肉生理特性和肌骨生物力学的驾驶员腰肌疲劳预测方法。 驾驶员腰部肌肉表面肌电信号能直接反映肌肉的疲劳信息。本文分析模拟静态驾驶、模拟动态驾驶和实车驾驶工况下驾驶员腰部肌肉收缩时的表面肌电信号特征变化，针对动态工况表面肌电信号易受动作干扰的问题，提出递归回归分析方法将表面肌电信号特征分解为静态特征和运动特征，以静态特征作为判断疲劳进程的依据。同时分析了多通道信号非独立、信号高通截止频率、窗宽和母小波种类等因素的影响，设计了带参数优化的支持向量分类器对模拟动态驾驶和实车驾驶样本进行分类，辨识精度分别达到93.04%和83.95%。 驾驶员-座椅界面分布载荷间接反映肌肉疲劳信息。为了降低动 态驾驶工况车辆振动和驾驶动作干扰，本文提出根据被试者乘坐位置将体压分布自适应分区，从座垫整体和局部提取不同区块单帧压力特征和帧间特征；分析驾驶员肌肉疲劳的体压分布总体影像特征，提出驾驶员腰部疲劳历程分为发展阶段、尝试调整阶段和确定调整阶段，并在随后的特征分析中进一步印证；最后设计带有序列浮动特征选择算法的高斯-贝叶斯分类器，分别对模拟动态驾驶和实车驾驶进行驾驶员腰肌疲劳分类辨识，平均分类正确率分别达到93.37%和88.282%。 驾驶员腰部受力复杂，且腰部浅层、深层椎旁肌群发出的运动单位动作电位混叠严重，直接建模条件不理想。为了最大程度降低干扰以直接分析肌肉疲劳特征，本文开展大鼠腓肠肌单根肌肉试验，分析不同神经激励程度和不同前负荷下肌肉力和肌电信号变化特征，然后根据肌肉运动单位激活、疲劳、恢复的现象规律，建立可疲劳腰部生物力学模型，提出以单根肌肉疲劳速率倒数为权值的腰部整体疲劳加权预测方法，以周围主要的8个肌群肌纤维状态表达静态和动态工况下腰部整体疲劳水平。用静动态预测结果对照文献实验值对模型进行验证，线性相关性分别达到0.9514、0.8246。
Lumbar muslce fatigue tends to be significant and becomes a major aspect affecting the vehicle comfort during long-distance driving. It threatens drivers’ spine health and driving safety. However, due to complicated reasons, drivers’ lumbar muscle fatigue is hard to detect and unable to provide support for the optimization design of the car seat comfort. In order to solve this problem, this paper constructed a dynamic identification algorithm of fatigue state based on the internal and external characteristics of drivers’ lumbar muscle. Meanwhile, a fatigue prediction method of drivers' lumbar muscle was established based on the muscle physiological characteristics and muscular biomechanics. The surface EMG signals of drivers’ muscle can directly reflect the muscle fatigue information. This paper analyzed the sEMG characteristics changes of drivers’ lumbar muscles in the condition of simulated static driving, simulated dynamic driving and road test. Bescause the sEMG signals could be easily disturbed by the driving movements, the paper proposed a method of recursive regression analysis decomposing the sEMG signal characteristics into static features and dynamic features and used static features as the basis for predicting fatigue progression. At the same time, the influences of channel correlation, cut-off frequency of high-pass filter, window width and mother wavelet were analyzed. Finally, a support vector classifier with parameter optimization was designed to classify the samples from dynamic driving and road test and the average classification accuracy reach 93.04% and 83.95% respectively. Driver-seat interface distribution pressure indirectly reflects muscle fatigue information. In order to reduce the disturbance of the vehicle vibration and driving motion under dynamic driving conditions, this paper proposed to adaptively partition the body pressure distribution according to the seating positions of the subjects. The single-frame and inter-frame pressure characteristics of different blocks from the seat cushion were extracted as a whole and locally. The general image characteristics of the body pressure distribution of muscle fatigue were analyzed. The paper found the fatigue history of driver's lumbar could be divided into development phase, tentative posture adjustment phase and determinate posture adjustment phase, which were further confirmed in the subsequent feature analysis. Finally, a Gaussian-Bayesian classifier with sequential floating feature selection algorithm was designed to classify the driver's lumbar fatigue in both simulated and actual vehicle driving. The average classification accuracy was 93.37% and 88.282% respectively. Because the driver's lumbar has complex loads and the action potentials of the motor units in the superficial and deep paraspinal muscles overlap seriously, the direct modeling conditions are not ideal. In order to directly analyze the muscle fatigue characteristics and minimize interferences, this paper carried out single muscle tests using gastrocnemius in rats and analyzed the changes of muscle force and myoelectric signal under different levels of nerve stimulation and preload. According to the activation, fatigue and recovery states of muscle movement units, a driver’s lumbar musculoskeletal biomechanical model with fatigue was established. The reciprocal of the single muscle fatigue rate was used as weights of total lumbar fatigue weighted prediction method. The overall lumbar fatigue level in static and dynamic conditions was predicted using the main eight muscle groups and verified by the experimental data from the literatures. The linear correlation coefficient of static and dynamic conditions reach 0.9514 and 0.8246 respectively.
