Song Zhi-huan

Improved PCA with optimized sensor locations for process monitoring and fault diagnosis

Process monitoring and fault diagnosis using the principal component analysis (PCA) has been studied intensively and applied to industry processes. The emphasis of most PCA-based works has been mainly on procedures to perform monitoring and diagnosis for given a set of sensors, and little attention is paid to the actual location of sensors for efficient detection and identification of process faults. In this paper, graph-based techniques are used to optimize sensor locations to obtain the maximum fault resolution. Based on the optimized sensor network, an improved PCA is proposed by introducing two new statistics of PVR and CVR to replace the Q statistic in the conventional PCA. The improved PCA can efficiently detect weak changes, and give an insight into the root cause of process faults. Simulation results of a CSTR process show that the improved PCA with optimized sensor locations is superior to the conventional methods.

Adaptive predictive control based on wavelet approximation models

Discusses approximate modelling and adaptive control of discrete time linear time-varying systems (LTVS). A novel model, the representation of LTVS in the wavelet time-frequency space, is established based on discrete wavelet transforms. The joint time-frequency properties of wavelet analysis are appropriate for describing the nonstationary characteristic of LTVS. In this model parameter estimation is fulfilled by means of some algebraic calculations. Finally, an adaptive predictive control scheme is proposed based on this representation. Simulation results demonstrate its good behaviour.

A novel NF-GMDH-IFL and its application to identification and prediction of nonlinear systems

The neurofuzzy GMDH with a feedback loop (NF-GMDH-FL) is presented, based on NF-GMDH and GMDH-FL. The feedback loop of NF-GMDH-FL are investigated and improved in this paper. The neuro fuzzy GMDH with an improved feedback loop (NF-GMDH-IFL) is developed, in which the redundant combinations and computations are discarded. NF-GMDH-IFL has faster training speed and less training time than NF-GMDH-FL. For the convenience of comparison, NF-GMDH, NF-GMDH-FL and NF-GMDH-IFL are applied to two examples, respectively. The simulations show that the capability of NF-GMDH-IFL is better than NF-GMDH and NF-GMDH-FL. Furthermore, the enormous time and memory are saved in NF-GMDH-IFL. The identification and prediction performance for nonlinear systems is observably improved.

Cross-directional modeling and control of paper-making processes in the discrete wavelet transform domain

This paper proposes a novel approach for cross-directional (CD) modeling and control of a paper machine based on discrete wavelet transforms. CD basis weight variations and the spatial shape of the slice lip are approximated at various resolutions, respectively. A new response model describing the relationship between the settings of the slice-screws to the basis weight profile is established in the discrete wavelet transform domain. Simulation results show that the size of the optimization and control problem associated with large dimensions can be significantly reduced.

Volterra series based predistortion for broadband RF power amplifiers with memory effects

Abstract RF power amplifiers (PAs) are usually considered as memoryless devices in most existing predistortiontechniques. However, in broadband communication systems, such as WCDMA, the PA memory effects are significant, and memoryless predistortion cannot linearize the PAs effectively. After analyzing the PA memory effects, a novelpredistortion method based on the simplified Volterra series is proposed to linearize broadband RF PAs withmemory effects. The indirect learning architecture is adopted to design the predistortion scheme and the recursiveleast squares algorithm with forgetting factor is applied to identify the parameters of the predistorter. Simulationresults show that the proposed predistortion method can compensate the nonlinear distortion and memory effectsof broadband RF PAs effectively.

GEKF, GUKF and GGPF based prediction of chaotic time-series with additive and multiplicative noises

On the assumption that random interruptions in the observation process are modelled by a sequence of independent Bernoulli random variables, this paper generalize the extended Kalman filtering (EKF), the unscented Kalman filtering (UKF) and the Gaussian particle filtering (GPF) to the case in which there is a positive probability that the observation in each time consists of noise alone and does not contain the chaotic signal (These generalized novel algorithms are referred to as GEKF, GUKF and GGPF correspondingly in this paper). Using weights and network output of neural networks to constitute state equation and observation equation for chaotic time-series prediction to obtain the linear system state transition equation with continuous update scheme in an online fashion, and the prediction results of chaotic time series represented by the predicted observation value, these proposed novel algorithms are applied to the prediction of Mackey–Glass time-series with additive and multiplicative noises. Simulation results prove that the GGPF provides a relatively better prediction performance in comparison with GEKF and GUKF.

The study of Naive Bayes algorithm online in data mining

Naive Bayes algorithm in data mining is studied online. The information and knowledge gained can be used for training new tuple online, which gives rise to the algorithm that can deal with huge amounts of database quickly. The program of the algorithm is also proposed. The practice usage of the algorithm shows its merit and efficiency.

Rough set based modeling and controller design in an internal model control system

This paper proposes framework of rough set (RS) theory based internal model control (IMC) system for coagulation control. Decision rules are extracted from RS information system for modeling and controller design. Correlation analysis method is used for time delay estimation. Discretization method based on Boolean reasoning is used for real value attributes coding. Matrix computation methods are used for rule extraction and dynamic modification. The problems of inconsistence and incompleteness are analyzed. Comparing with neural network (NN) methodology, the properties of RS based modeling are discussed. Simulations show that RS theory is a valuable technique for IMC.This paper proposes framework of rough set (RS) theory based internal model control (IMC) system for coagulation control. Decision rules are extracted from RS information system for modeling and controller design. Correlation analysis method is used for time delay estimation. Discretization method based on Boolean reasoning is used for real value attributes coding. Matrix computation methods are used for rule extraction and dynamic modification. The problems of inconsistence and incompleteness are analyzed. Comparing with neural network (NN) methodology, the properties of RS based modeling are discussed. Simulations show that RS theory is a valuable technique for IMC.

Statistical process monitoring with measured data corrupted by noise and gross error

Principal component analysis (PCA) is an efficient method to extract relationships between correlated variables and thus has been widely applied to the multivariate statistical process monitoring. However, the validity of the PCA is highly depending on the quality of measured process data which is usually contaminated by noises and gross errors in practice. In this paper, an improved PCA is presented to minimize the influences of corrupted data. The original measured data is first processed using conventional PCA to partially eliminate noises by abandoning principal components with small eigenvalues. Then the retained principal components are decomposed and rectified online by boundary-corrected wavelets, combined with techniques of shift-invariant transform and median filtering. The monitoring results of a simulated binary distillation column show that the proposed method has superior performance and more robust to noises and gross errors than conventional methods.

A fault diagnosis expert system for hydraulic system of injection moulding

Injection moulding is a kind of highly automatic machine. The hydraulic system is an important part for its proper operation. But because of the frequent faults, the productivity of injection moulding is significantly decreased. Traditionally, the expert system for fault diagnosis is only based on expert experiences. However, considering the complexity of faults of the hydraulic system, it sometimes leads to inaccurate results, even no results. We present a method based on the combination of an experts experiences and the objects structure and function. By using shallow and deep knowledge synthetically, the framework of an efficacious system is proposed. We describe the generalized knowledge base and the new diagnosis strategies and reference mechanism. We also give some practical rules in LISP language. In most cases, especially unexpected ones, this method can lead to better results.