<p>Electrical machines are widely used in industrial applications. Induction motors are the most commonly used electrical machines. The healthy operation of induction motors is important for the continuous operation of industrial processes. Therefore, the health status of induction motors is continuously monitored by collecting signals from various sensors to detect faults at early stages. Signals such as vibration, current, torque, voltage, and sound are commonly used to detect faults of induction motors. In this study, vibration, current, and torque signals were used to detect the bearing faults of a 1.5&#xa0;kW three-phase induction motor. The bearing faults were artificially implemented by using the electroplating method. In the proposed study, vibration, current, and torque signals were converted into spectrograms by using Short Time Fourier Transform (STFT) with/without Hilbert Transform (HT) and used as input in a Convolution Neural Network (CNN) model. Various combinations of spectrograms of input signals (three-phase current, three-axis vibration, and torque signals) were merged and used as input to the proposed CNN model. A graphical user interface (GUI) module was also designed in the LabVIEW (National Instruments) environment. The GUI module may be used to activate the input data channels manually in order to test the performance of the proposed method with the missing data or failed sensors. A highest accuracy rate of 99.75% was achieved by fusing the spectrograms representing three-axis vibration, three-phase current, and torque signals. The experimental results show that the proposed method can be used successfully for detecting the bearing faults of induction motors.</p>