齿轮是机械传动的关键零部件，齿轮的磨削烧伤会影响其传动的稳定性和疲劳寿命。通过材料中马氏体含量的变化判定齿轮的烧伤程度，利用磁巴克豪森噪声检测装置对激光模拟烧伤齿轮进行测试，采集烧伤齿面的巴克豪森噪声信号；通过巴克豪森噪声信号的包络与切向磁场的关系曲线提取特征值（包括峰值，峰值位置，半峰宽），利用自适应模糊神经网络进行训练，建立材料马氏体含量的预测模型。试验结果表明，该方法具有检测和表征微观金相组织中马氏体深度的能力，同时可以避免激励频率对巴克豪森噪声信号输出的影响。通过拟合优度参数R²=0.964 1和均方根误差R_MSE=16.981 7进一步验证了自适应模糊神经网络模型具有较高的准确性，可用于预测齿轮烧伤的程度。

Gears are the key components of mechanical transmission. The grinding burn affects the stability and fatigue life of gears. This paper determines the burn degree of gear burn by tracking the change of microstructure of material. The laser simulated burn gear was tested by the magnetic Barkhausen noise detector, and the Barkhausen noise signal of the burned tooth surface is collected. The characteristic value of the signal is extracted by the relationship between the envelope of the Barkhausen noise signal and the tangential magnetic field (including peak, peak position, half-width), and a prediction model for the evaluation of martensite content is established by using adaptive fuzzy neural network for training. The experimental results show that the method has the ability to detect and characterize the martensitic depth in microscopic metallographic structure, and can avoid the influence of excitation frequency on the output of Barkhausen noise signal. The fit parameter R²=0.964 1 and mean square root error R_MSE=16.981 7 verifies accuracy of the adaptive fuzzy neural network model and can accurately obtain the martensite depth information of the burn gear, and then predict the degree of gear burn.