Additive manufacturing (AM), also known as 3D printing, is a disruptive technique and provides good compensation for conventional manufacturing methods. In AM, 3D parts are processed in a layer-by-layer manner following the designed 3D model and toolpaths. The rapid advancement of AM allows for an unprecedented design freedom for manufacturing complex, composite, and hybrid structures with high precision, which cannot be achieved using traditional fabrication routes. However, the AM process development and optimization usually requires costly and time-consuming trial-and-error experiments, thereby limiting the further application of AM. Machine learning (ML), as a new type of artificial intelligence technology, can accelerate the research and development in many aspects of AM; therefore AM has received extensive attention from academia and industry. With the assistance of ML, AM can be expedited and well optimized. Moreover, the relationship between the process parameters and achievable property of the alloys can be well revealed through ML, which is difficult using conventional methods. The ML technique has exhibited promising potentials in accumulating process optimization and novel alloy design for AM recently. Hence, this work reviews the research progress of the ML-assisted AM in the past decade.

In this paper, first, the ML technology used in AM is described. In general, ML methods can be divided into supervised learning, unsupervised learning, semisupervised learning, and reinforcement learning. According to studies, each ML method has many applications. Therefore, the typical applications for each ML method are introduced (Fig. 1). Second, the application of ML in the control and optimization of the AM metal materials, including the process monitoring and quality control, prediction of the process window, and optimization of the deposition toolpath, is discussed. By combining appropriate ML methods, the AM development processes can be considerably expedited and quality of the deposited parts can be stabilized. Third, the status of research and application of ML in the development of new alloys for AM is introduced. The correlative applications mainly include alloy composition design and prediction of microstructure and property of the deposited alloys. Recent years witnessed the growing research interests in the development of novel alloy materials used for AM (Fig. 8). Because it has been demonstrated that ML is an efficient way to accelerate the development period of novel alloy materials. With more available data accumulation, it can be expected that ML will have a broad prospect in novel alloy development for AM, which could create high-performance alloys for harsh industrial applications.

With the development of artificial intelligence and computer science, ML has been widely used in AM in recent years. The combination of ML and AM avoids a large quantity of trial-and-error costs, thereby reducing the development period of the AM. This work reviews the progress of machine learning-based AM process optimization and the novel alloy materials developments. The application of ML in the control and optimization of the AM includes the process monitoring, quality control, prediction of the process window, and optimization of the deposition toolpath (Fig. 10). The research and application of ML in the development of novel alloy materials based on AM include alloy composition design, microstructure, and property prediction. Finally, the future development trends of ML in the AM were outlined. In studies, the ML method usually focuses on a particular phase of the AM, which considerably limits the application and promotion of machine learning. The development of the generic ML algorithm for AM will further promote the application of ML in AM, which is also the critical research direction of machine learning-assisted AM in the future. For ML-based novel alloy materials developments in AM, several studies have shown that ML can effectively avoid the high costs of the traditional trial-and-error methods. However, ML requires a large number of databases to train the model. Therefore, the construction and development of an effective database is the precondition for ML. In recent years, a large amount of literature regarding AM of metallic materials has been published, which means a large amount of experimental data has been accumulated, and this is the fundamentals for the development of ML technology. With the development of practical data mining technology, the vast database will promote the development of novel metal materials for AM.