Predictive maintenance on rotating machinery, reducing downtime, and averting catastrophic failures all depend on ball bearing condition monitoring. Traditional techniques use accelerometers to collect real-world vibration signals, but collecting large and balanced datasets across fault types is often challenging due to time, cost, and safety concerns. This paper presents an artificially generated data-based machine learning method for ball bearing fault diagnosis. Numerical simulation techniques are used to reproduce vibration signals during normal operation, inner race fault, outer race fault, and ball defect. The synthetic signals are checked against benchmark datasets to ensure physical validity. Classifiers such as Random Forest, Support Vector Machine (SVM), and Deep Neural Networks (DNN) are trained using time, frequency, and time-frequency features derived from the synthetic signals. Experiments on the validation of the CWRU and pad-born datasets demonstrate that models trained on synthetic data achieve over 97% accuracy and generalize well to actual signals, with over 92% accuracy. The study demonstrates the potential benefits of machine learning based on synthetic data for precise condition monitoring, especially when labeled data is difficult to obtain.