Fault Diagnosis Of Shaft Furnace Roasting Processes Using Concurrent Projection To Latent Structures

The shaft furnace roasting process is one of the most important units in the mineral processes, and its main technical process is to carry out a high-temperature deoxidizing-magnetic roasting to the crude ores, and to transform the weak magnetic ore into strongly magnetic ore, then to the next process to obtain high grade ore concentrate. The product quality is directly related to whether the production index which measures the effect of ore selection, such as the mental recovery rate and the grade of finished ores. Therefore it is necessary to diagnose the faults, especially the ones related to the product quality for the shaft furnace roasting processes.The shaft furnace roasting process contains complex physical and chemical reduction change and a lot of uncertain factors, leading to large production fluctuation. If working conditions abnormally change, or the operator cannot react properly or timely, it may lead to the unstable working state or even stagnation. Some unexpected faulty situations, e.g. fire-emitting, flame-out, under-reduction and over-reduction may happen. These faults are closely related to the safety working of process and product quality, i.e. the magnetic tube recovery rate (MTRR). Fault diagnosis of shaft furnace roasting processes is therefore intensively studied. But The traditional fault diagnosis methods encounter the following issues:i) rule-based reasoning method can only diagnose faults caused by the improper loop set points, i.e. the temperature of combustion chamber, the flow rate of reducing gas, and the ore discharging time, but cannot diagnose which variable leads to the faults; and ii) data-driven principal component analysis (PCA) based fault diagnosis method is cannot diagnose the relations between the faults and the product quality.In view of the above issues, this thesis focuses on data driven fault diagnosis of shaft furnace roasting processes. The work is partly supported by the National 973 project and National Natural Science Foundation project. The research work mainly includes the following three aspects:(1) Data-driven concurrent projection to latent structures (CPLS) based fault diagnosis approach is proposed, in order to monitor the process and diagnose the variables which lead to the faults. First, this method utilizes the historical normal data to perform concurrent projection to latent structures decomposition to have the CPLS model. Then comprehensive fault detection is achieved using the fault detection indices defined for the variations in the partitioned subspaces. Using the CPLS model, this proper proposes a CPLS-contribution plots based fault diagnosis method to diagnose the faulty variables. Finally, the experiment results on Tennessee-Eastman process demonstrate the effectiveness of the proposed methods.(2) The proposed CPLS based fault methods are applied to the shaft furnace roasting processes, in order to monitor the process and determine whether the fault influence the product quality of the shaft furnace roasting process. The proposed CPLS based contribution plots method is applied to diagnose the faulty variables. In order to further distinguish each fault of shaft furnace roasting process, traditional reconstruction based method is applied.(3) Experiments on the shaft furnace roasting process are carried out on a simulation platform. Under normal working conditions of roasting process, results of fault diagnosis experiment show that the proposed methods in this paper decrease the false positive rate by 5.3% compared with the PCA based fault diagnosis method. Experimental results in abnormal working conditions of roasting process demonstrate the following advantages of the proposed methods:i) the comprehensive fault detection method in this paper not only detects fault condition in roasting process, but also displays whether this fault condition affects the MTRR or not; ii) for the over-reducing abnormal condition, the PCA based method only identifies the flow rate of reducing gas, while the proposed CPLS based contribution plots further pinpoint the abnormal temperature of combustion chamber. The experiment results demonstrate that the CPLS contribution plots method is more effective to diagnose the faulty variable of quality-relevant faults.