Aiming at the problem of low diagnostic accuracy caused by difficult extraction of motor rolling bearing faults a feature extraction method of local characteristic-scale decomposition LCD multi-scale fusion was proposed. The extracted feature vectors were input into the support vector machine SVM optimized by golden jackal optimization GJO for motor bearing fault diagnosis so as to improve the accuracy of fault diagnosis. Methods Firstly the LCD algorithm was used to decompose motor bearing signals collected by the vibration sensor to obtain intrinsic scale components ISC of the signal at the first three scales. Secondly the energy ratio and energy entropy of each ISC were calculated to obtain 6 sets of feature vectors. Then the feature vector matrix was constructed and randomly selected according to 3 ∶ 2 to generate a training set and test set. Finally the training set was input into the golden jackal algorithm optimization support vector machine GJO-SVM and the GJO algorithm was used to optimize the penalty parameter C and kernel parameter g in the support vector machine to establish the GJO-SVM fault diagnosis model. The test set was input into the model to realize the fault identification of the bearing. Results Through simulation experiments it is verified that the feature extraction method of LCD multi-scale fusion features can effectively extract the fault information of motor bearings and the fault diagnosis accuracy of GJO-SVM is 97. 86%. Conclusion Under the same conditions the diagnostic accuracy of the LCD multi-scale fusion feature is improved by 1. 79% compared with the fusion feature of variational modal decomposition VMD . Compared with the particle swarm optimization-support vector machine PSO- SVM method and the fruit fly optimization algorithm-support vector machine FOA-SVM method the GJO-SVM method has a higher fault identification effect for diagnosing motor bearing faults.