Research On Incipient Fault Detection, Prediction And Maintenance Methods For Hydraulic Tube Tester

As one type of fluid transporting pipeline, steel tube, which is an essential and vital steel product in various fields of the national economy, has attracted spread interests in its quality, relevant techniques for its quality detection and innovation of the relevant equipments in particular. Great progress has been achieved in these fields. Being a necessary part of quality inspection during the production process of steel tube, pressure test plays an important role in the assessment of steel tube. Hydraulic tube tester, which is especially designed and produced in order to test the pressure resistance ability of steel tube, is a complex system composed of mechanics, hydraulics and electronic control, et al. On one hand, as the hydrostatic testing process is a critical quality inspection procedure in the production of steel tube, the running status of hydraulic tube tester has a direct effect on production efficiency of steel tube. On the other hand, as a complex high-press system, the failure of hydraulic tube tester will not only seriously affect the safety and reliability of process operation, but even result in huge loss of personnel and property. Therefore, it is significant to study feasible real-time monitoring, fault detection and prediction methods to support reliable basis for production management, which will improve the performance of machine detection and guarantee the safety and reliability of production.Based on the aforementioned consideration, by setting the3#hydraulic tube tester in the Baosteel Tube Branch, imported from the DEMAG Compony, Germany in1995, as the research object, this dissertation further studies its work mechanism and thus developes a series of incipient fault detection, prediction and maintenance methods focusing on the characteristics of incipient fault occurred in the hydraulic system of this type of hydraulic tube tester, which shows important practical values.(1). The boost system and balance system are simplified to construct their corresponding mathematical models after the operation mechanism of the main hydraulic system in the hydraulic tube tester has been summarized and analyzed. Based on this, some submodels concerning fault simulation are developed for the main subsystems, such as oil supply system, boost system, balance system and so on. Finally, the whole simulation model is developed for the main hydraulic system of hydraulic tube tester based on the AMEsim software, which realizes the dynamlic simulation of this hydraulic system under normal and commonly faulty conditions, respectively. According to the simulation results of this hydraulic system, the incipient faults can be divided into two categories, i.e. incipient fault with and without periodic fault symptom.(2). For the first type of incipient fault with periodic fault symptom, taking single piston loose shoes fault of hydraulic pump in the hydraulic tube tester for example, by selecting a characteristic variable, an early fault detection algorithm is proposed by a combination of Chaos oscillator and sliding window symbol sequence statistic methods. The proposed method firstly chooses Chaos oscillator to diagnose this type of incipient fault based on the insight that the phase transition of Duffing oscillator is very sensitive to a periodic weak signal and immune against to other noises. Then, considering that it is not easy for computer to discriminate the intermittent chaos phenomenon, sliding window symbol sequence statistics method is developed to realize online real-time diagnosis, and critical parameter configuration are analyzed for window size and depth of symbol tree.(3). For the second type of incipient fault without periodic fault symptom, which is notrealistic to effectively detect the connotative fault symptom through a single process variable, a probabilistic fault prediction method is developed based on multivariate statistical analysis technique. The proposed method firstly unfolds three-way batch process data into a two dimensional matrix, applies principle component analysis (PCA) to this two dimensional matrix to extract corresponding statitistical index reflecting the status of the running system, i.e. the combined index. Then, sliding window cumulative sum mentod is proposed to decrease the stochastic volatility, and increase the regularity buried in each statistical time-series. After that, Bayesian AR model is used to predict the transformed index to get its probability distribution at future time, and calculates its corresponding fault probability and time to failure.(4). Based on the analysis of system’s or equipment’s deterioration process, for the first typeof incipient fault, a predictive maintenance method is developed by identifying the mode change point according to the ratio of two time spans, which are the duration of chaos motion and large periodic motion of the Duffing oscillator in a cycle. For the second type of incipient fault, the proposed method partitions the system statuses into3statuses, i.e. normal status, critical statuse and fault status. Then, it calculates the reliability according to the fault probability, and realizes predictive maintenance by determinining the deterioration mode and reliability of current system.Finally, the potential further research direction in the area of incipient fault detection and prediction is discussed after summarizing the whole work in this thesis.