This paper provides a comprehensive review and roadmap for intelligent fault diagnosis (IFD) using machine learning theories. IFD aims to automate the process of recognizing the health states of machines, reducing the reliance on human labor and improving diagnostic accuracy. The review is structured into three main sections: past, present, and future. 1. **Past: IFD using traditional machine learning theories** - **Data Collection**: Sensors are used to collect data from various sources such as vibration, acoustic emission, temperature, and current. - **Artificial Feature Extraction**: Features are extracted from the collected data, and feature selection methods are used to identify the most relevant features. - **Health State Recognition**: Diagnosis models are trained to recognize the health states of machines based on the selected features. Approaches include expert systems, artificial neural networks (ANN), support vector machines (SVM), and other methods. 2. **Present: IFD using deep learning theories** - **Big Data Collection**: Deep learning-based approaches handle large volumes of data more effectively than traditional methods. - **Deep Learning-Based Diagnosis**: Approaches include stacked autoencoders (AE), deep belief networks (DBN), convolutional neural networks (CNN), and ResNet. These methods automatically learn fault features from raw data, providing end-to-end diagnosis models. 3. **Future: IFD using transfer learning theories** - **Motivation**: Transfer learning aims to overcome the challenges of limited labeled data in engineering scenarios by applying knowledge from one or multiple tasks to new tasks. - **Definitions**: Transfer learning in IFD involves transferring knowledge from one task to another, addressing issues like feature-based, GAN-based, instance-based, and parameter-based approaches. - **Categories**: Various categories of transfer learning-based approaches are discussed, each with its own advantages and challenges. The paper concludes with a discussion on future challenges and a roadmap for IFD, emphasizing the importance of combining deep learning with transfer learning to enhance the applicability of IFD in real-world engineering scenarios.This paper provides a comprehensive review and roadmap for intelligent fault diagnosis (IFD) using machine learning theories. IFD aims to automate the process of recognizing the health states of machines, reducing the reliance on human labor and improving diagnostic accuracy. The review is structured into three main sections: past, present, and future. 1. **Past: IFD using traditional machine learning theories** - **Data Collection**: Sensors are used to collect data from various sources such as vibration, acoustic emission, temperature, and current. - **Artificial Feature Extraction**: Features are extracted from the collected data, and feature selection methods are used to identify the most relevant features. - **Health State Recognition**: Diagnosis models are trained to recognize the health states of machines based on the selected features. Approaches include expert systems, artificial neural networks (ANN), support vector machines (SVM), and other methods. 2. **Present: IFD using deep learning theories** - **Big Data Collection**: Deep learning-based approaches handle large volumes of data more effectively than traditional methods. - **Deep Learning-Based Diagnosis**: Approaches include stacked autoencoders (AE), deep belief networks (DBN), convolutional neural networks (CNN), and ResNet. These methods automatically learn fault features from raw data, providing end-to-end diagnosis models. 3. **Future: IFD using transfer learning theories** - **Motivation**: Transfer learning aims to overcome the challenges of limited labeled data in engineering scenarios by applying knowledge from one or multiple tasks to new tasks. - **Definitions**: Transfer learning in IFD involves transferring knowledge from one task to another, addressing issues like feature-based, GAN-based, instance-based, and parameter-based approaches. - **Categories**: Various categories of transfer learning-based approaches are discussed, each with its own advantages and challenges. The paper concludes with a discussion on future challenges and a roadmap for IFD, emphasizing the importance of combining deep learning with transfer learning to enhance the applicability of IFD in real-world engineering scenarios.