The cutters of shield cutterhead are prone to damage due to complex geologies encountered, and the existing cutter fault diagnosis methods cannot comprehensively and accurately judge the cutter conditions. Therefore, the principles and characteristics of machine learning algorithms such as support vector machine(SVM), back propagation neural network(BPNN), extreme learning machine(ELM), and random forest(RF) are analyzed, the principle and characteristics of common faults such as wear overrun, stuck rotation, and eccentric wear of the disc cutter in atmospheric shield cutterhead are examined, and a fault diagnosis model of atmospheric shield cutterhead disc cutter based on machine learning is designed to conduct hierarchical fault diagnosis. The temperature, instantaneous speed ratio, and average speed ratio that are more sensitive to the disc cutter failure are selected, and the shield tunneling field data are used to conduct experiments, so as to determine whether the disc cutters fail. The accuracies of the firstlayer SVM, BPNN, ELM, and RF algorithm models reach 87.49%, 88.69%, 78.27%, and 96.45%, respectively. When the disc cutters fail, the accuracies of the secondlayer SVM, BPNN, ELM, and RF algorithm models reach 90.24%, 86.76%, 79.41%, and 97.06%, respectively. These validate the rationality and effectiveness of the designed machine learningbased atmospheric shield cutterhead disc cutter fault diagnosis model. It is found that the RF algorithm has a high accuracy in judging whether the disc cutters fail and the fail type. Thus, the cost of enterprise construction and tool change can be effectively reduced, and the tunneling efficiency of the shield can be improved.