Backgrounds <p>The occurrence conditions of underground coal seams are complex, and the cutting operation environment is harsh. When the hardness of coal-rock suddenly changes, the mismatch between cutting motion parameters and coal-rock hardness generates severe impact loads, which poses a serious threat to the safe operation of tunneling equipment.</p> Purposes <p>The intelligent operation of tunneling equipment requires a certain degree of adaptive cutting capability. This paper studies an adaptive cutting method based on the coal-rock hardness range recognition.</p> Methods <p>Collect multi-sensor information including cutting motor current, driving cylinder torque, and vibration of the cutting head, construct multi-domain time–frequency-entropy-energy features, and utilize deep belief network (DBN) to present a coal-rock hardness range recognizer. Adjust the cutting motion parameters adaptively based on the identified coal-rock hardness changes.</p> Results <p>Compared with traditional feature dimension reduction methods, DBN can learn abstract feature information that is more conducive for the coal-rock hardness classification. Under the condition of sample scarcity, DBN has higher recognition accuracy and generalization performance compared to traditional recognition methods. The feasibility of our proposed method was verified using the constructed coal-rock cutting simulation test platform.</p> Conclusions <p>The proposed DBN based coal-rock hardness range recognition method does not require manual feature screening and secondary dimension reduction, and the recognition effect is better than those existing methods. Ground simulation tests showed that the roadheader can adaptively adjust the cutting motion parameters based on the identified coal-rock hardness change, effectively improving the operating status of the roadheader.</p>