The process faults of shaft furnace roasting processes, e.g. fire-emitting, flame-out, under-reduction, and over-reduction are undesirable for stable operation of the processes. The processes share multiple complexities such as multi-variate and strong correlations, which make it difficult to diagnose the faults using model-based or knowledge-based methods. In this paper, a data-driven fault diagnosis method for shaft furnace roasting processes is presented based on reconstruction and reconstruction-based contribution. The proposed method exploits historical faulty data to derive fault directions to identify ongoing faults with the help of additional explanation from contribution plots. A case study on a simulation system of shaft furnace roasting processes illustrates the effectiveness of the proposed method.The process faults of shaft furnace roasting processes, e.g. fire-emitting, flame-out, under-reduction, and over-reduction are undesirable for stable operation of the processes. The processes share multiple complexities such as multi-variate and strong correlations, which make it difficult to diagnose the faults using model-based or knowledge-based methods. In this paper, a data-driven fault diagnosis method for shaft furnace roasting processes is presented based on reconstruction and reconstruction-based contribution. The proposed method exploits historical faulty data to derive fault directions to identify ongoing faults with the help of additional explanation from contribution plots. A case study on a simulation system of shaft furnace roasting processes illustrates the effectiveness of the proposed method.The process faults of shaft furnace roasting processes, e.g. fire-emitting, flame-out, under-reduction, and over-reduction are undesirable for stable operation of the processes. The processes share multiple complexities such as multi-variate and strong correlations, which make it difficult to diagnose the faults using model-based or knowledge-based methods. In this paper, a data-driven fault diagnosis method for shaft furnace roasting processes is presented based on reconstruction and reconstruction-based contribution. The proposed method exploits historical faulty data to derive fault directions to identify ongoing faults with the help of additional explanation from contribution plots. A case study on a simulation system of shaft furnace roasting processes illustrates the effectiveness of the proposed method.The process faults of shaft furnace roasting processes, e.g. fire-emitting, flame-out, under-reduction, and over-reduction are undesirable for stable operation of the processes. The processes share multiple complexities such as multi-variate and strong correlations, which make it difficult to diagnose the faults using model-based or knowledge-based methods. In this paper, a data-driven fault diagnosis method for shaft furnace roasting processes is presented based on reconstruction and reconstruction-based contribution. The proposed method exploits historical faulty data to derive fault directions to identify ongoing faults with the help of additional explanation from contribution plots. A case study on a simulation system of shaft furnace roasting processes illustrates the effectiveness of the proposed method.The process faults of shaft furnace roasting processes, e.g. fire-emitting, flame-out, under-reduction, and over-reduction are undesirable for stable operation of the processes. The processes share multiple complexities such as multi-variate and strong correlations, which make it difficult to diagnose the faults using model-based or knowledge-based methods. In this paper, a data-driven fault diagnosis method for shaft furnace roasting processes is presented based on reconstruction and reconstruction-based contribution. The proposed method exploits historical faulty data to derive fault directions to identify ongoing faults with the help of additional explanation from contribution plots. A case study on a simulation system of shaft furnace roasting processes illustrates the effectiveness of the proposed method.
