非线性时序模型在机械系统故障诊断中的应用

提出一种基于非线性自回归时间序列模型(gereral expression for linear and nonlinear auto-regressive model,简称GNAR模型)的机械系统状态识别与故障诊断方法.利用采集系统工作过程中的特征信号建立GNAR模型；用主成分分析策略生成模型特征量,对训练样本的特征量进行识别和分类,得到各种参考模式；将几何距离判别函数作为状态分类的原则,根据待判系统特征量与各类参考模式的Euclide距离进行状态识别和故障判别.对车床颤振试验数据及高速离心空气压缩机故障数据的分析表明,该方法快捷、高效,诊断成功率较好,具有良好的工程应用前景.     

基于小波网络的时间序列预报及其应用

对关键特征参数的准确预报是实现故障预报的重要环节 。本文提出了基于小波网络的非线性时间序列预报模型，探讨了非线性序列预报在故障预报中的应用。从函数逼近角度给出了这种模型的理论基础，给出了小波网络的学习算法，最后给出在空间推进系统上的应用实例。     

旋翼振动监测信号的演化分析与损伤跟踪方法

基于非线性涡流(nonlinear eddy current,简称NEC)检测技术搭建了实验系统,对Q195碳素钢和304奥氏体不锈钢两种常用核电结构材料的塑性损伤程度进行无损定量评价研究。发现材料的塑性损伤程度与非线性涡流检测信号频谱图中基频幅值、三次谐波幅值存在一定线性关系。不同材料的线性关系存在差异,Q195碳素钢的检测信号随损伤程度增大而下降,304奥氏体不锈钢的检测信号随损伤程度增大而上升。通过开发实验系统、进行塑性变形导入和非线性涡流检测实验,分析检测信号与塑性变形程度的相关性,发现检测信号中基波幅值及三次谐波幅值与检测试件的塑性变形程度具有良好相关性,验证了本研究方法对两种典型核电结构材料塑性变形无损定量评价的有效性与可行性。     

Nonlinear modeling and identification of the electro-hydraulic control system of an excavator arm using BONL model

Electro-hydraulic control systems are nonlinear in nature and their mathematic models have unknown parameters. Existing research of modeling and identification of the electro-hydraulic control system is mainly based on theoretical state space model, and the parameters identification is hard due to its demand on internal states measurement. Moreover, there are also some hard-to-model nonlinearities in theoretical model, which needs to be overcome. Modeling and identification of the electro-hydraulic control system of an excavator arm based on block-oriented nonlinear（BONL） models is investigated. The nonlinear state space model of the system is built first, and field tests are carried out to reveal the nonlinear characteristics of the system. Based on the physic insight into the system, three BONL models are adopted to describe the highly nonlinear system. The Hammerstein model is composed of a two-segment polynomial nonlinearity followed by a linear dynamic subsystem. The Hammerstein-Wiener（H-W） model is represented by the Hammerstein model in cascade with another single polynomial nonlinearity. A novel Pseudo-Hammerstein-Wiener（P-H-W） model is developed by replacing the single polynomial of the H-W model by a non-smooth backlash function. The key term separation principle is applied to simplify the BONL models into linear-in-parameters struc~tres. Then, a modified recursive least square algorithm（MRLSA） with iterative estimation of internal variables is developed to identify the all the parameters simultaneously. The identification results demonstrate that the BONL models with two-segment polynomial nonlinearities are able to capture the system behavior, and the P-H-W model has the best prediction accuracy. Comparison experiments show that the velocity prediction error of the P-H-W model is reduced by 14%, 30% and 75% to the H-W model, Hammerstein model, and extended auto-regressive （ARX） model, respectively. This research is helpful in controller design, system monitoring and diagnosis.     

Nonlinear modeling of compliant mechanisms incorporating circular flexure hinges with finite beam elements

Modeling of compliant mechanisms incorporating flexure hinges is mainly focused on linear methods. However, geometrically nonlinear effects cannot be ignored generally. This work shows that nonlinear behavior plays an important role in the deformation and stress analysis, which consequently impacts the design of compliant mechanisms. In this study a nonlinear higher order finite beam element based modeling approach is presented strongly reducing the computation time of nonlinear models. Planar deformation and mechanical stress of a single circular flexure hinge under a wide range of loads is modeled and computed with the proposed approach. A comparison with a 3D-nonlinear finite element model shows very good agreement and validates the beam model. It is shown that the linear and nonlinear deformation behavior of a single flexure hinge deviate marginally so that linear modeling approaches are sufficient. Furthermore a planar displacement amplification mechanism incorporating circular flexure hinges is studied by means of the same method highlighting the distinct deviation of the behavior of the geometrically nonlinear model from its linear prediction. In conclusion the nonlinear behavior at the system level can not longer be neglected. Finally, a study shows that different designs of the displacement amplification mechanism are achieved when linear or nonlinear modeling approaches are applied.     

Nonlinear predictive control for Hammerstein–Wiener systems

This paper discusses a nonlinear Model Predictive Control (MPC) algorithm for multiple-input multiple-output dynamic systems represented by cascade Hammerstein–Wiener models. The block-oriented Hammerstein–Wiener model, which consists of a linear dynamic block embedded between two nonlinear steady-state blocks, may be successfully used to describe numerous processes. A direct application of such a model for prediction in MPC results in a nonlinear optimisation problem which must be solved at each sampling instant on-line. To reduce the computational burden, a linear approximation of the predicted system trajectory linearised along the future control scenario is successively found on-line and used for prediction. Thanks to linearisation, the presented algorithm needs only quadratic optimisation, time-consuming and difficult on-line nonlinear optimisation is not necessary. In contrast to some control approaches for cascade models, the presented algorithm does not need inverse of the steady-state blocks of the model. For two benchmark systems, it is demonstrated that the algorithm gives control accuracy very similar to that obtained in the MPC approach with nonlinear optimisation while performance of linear MPC and MPC with simplified linearisation is much worse.     

基于状态反馈线性化的阀控非对称缸系统模型预测控制

针对使用PID方法对阀控非对称液压缸位置控制中出现的超调问题,以及传统非线性模型预测控制优化求解计算时间较长的问题,提出了一种基于状态反馈线性化的阀控非对称缸模型预测控制方案。首先建立了阀控系统状态空间模型,运用微分几何理论讨论系统可反馈线性化的充要条件,并将非线性系统映射为新坐标空间内的线性系统模型;设计了反馈线性化模型预测控制器(Feedback Linearization Model Predictive Controller, FLMPC),讨论了线性系统下的约束问题,其中由于系统仿真预测时域远小于系统响应时间,对模型预测控制的损失函数加以修正。结果证明,在相同输入情况下,反馈线性化系统与原系统的位置误差满足控制需要,且在保证被控对象快速稳定控制的条件下,对比该算法与非线性模型预测控制的单步计算时间,证明该算法能够缩短计算时间。     

Hybrid modeling for robust nonlinear multivariable control

This paper describes a hybrid modeling approach and compares it to classic linear dynamic models and nonlinear models. Particular attention is given to the performance of each type of model when embedded in a multivariable model predictive control system. The hybrid approach combines linear state-space model with a nonlinear neural network correction. Confidence computations are used to determine the amount of correction applied. The combined model is adapted online to address changes in process operating range. The hybrid structure offers several benefits from a control perspective. It is evolutionary, building on the rich theoretical foundation of linear model predictive control. It can model nonlinear processes. It adapts online. When compared to other linear and nonlinear modeling techniques for control purposes, it has several specific advantages that make it ideally suited to particular applications. These applications include modeling and controlling: nonlinear processes, processes with slowly changing inputs, processes with interacting variables, and small systems with fast cycle time requirements.     

Wavelet spectrum analysis approach to model validation of dynamic systems

Feature-based validation techniques for dynamic system models could be unreliable for nonlinear, stochastic, and transient dynamic behavior, where the time series is usually non-stationary. This paper presents a wavelet spectral analysis approach to validate a computational model for a dynamic system. Continuous wavelet transform is performed on the time series data for both model prediction and experimental observation using a Morlet wavelet function. The wavelet cross-spectrum is calculated for the two sets of data to construct a time-frequency phase difference map. The Box-plot, an exploratory data analysis technique, is applied to interpret the phase difference for validation purposes. In addition, wavelet time-frequency coherence is calculated using the locally and globally smoothed wavelet power spectra of the two data sets. Significance tests are performed to quantitatively verify whether the wavelet time-varying coherence is significant at a specific time and frequency point, considering uncertainties in both predicted and observed time series data. The proposed wavelet spectrum analysis approach is illustrated with a dynamics validation challenge problem developed at the Sandia National Laboratories. A comparison study is conducted to demonstrate the advantages of the proposed methodologies over classical frequency-independent cross-correlation analysis and time-independent cross-coherence analysis for the validation of dynamic systems.     

弹性体非线性振动系统参数辨识的一种时域法

提出了对弹性体非线性振动系统参数辨识与预测的一种时域模型法。它可视为时间序列分析中的 AR模型法在非线性领域内的一种推广。该方法首先将非线性振动系统中的非线性恢复力和非线性阻尼力用某一函数级数(例如幂级数 )表示 ,然后 ,先用线性模型来逼近原系统 ,应用线性系统辨识方法确定系统阶次 ,再确定系统中非线性恢复力、阻尼力的结构 ,建立非线性模型并辨识各项参数 ,最后进行预测。算例表明 ,用该方法建立的模型能够较好地反映系统的非线性特性 ,并能提高模型预测的准确性。     

Multivariate calibration for machine health monitoring: kernel partial least squares combined with variable selection

This work proposes and demonstrates the use of data mining techniques for machine health monitoring through a multivariate calibration model. It utilizes a genetic algorithm (GA)-based variable selection combined with a preprocessing technique of orthogonal signal correction (OSC) for constructing reliable calibration models of shaft misalignment conditions. Improper aligning of shafts often leads to severe problems in many rotating machines. Thus the prediction of shaft alignment conditions is quite essential in making decisions on when to perform alignment maintenance. The main goal of this calibration model is to predict misalignment conditions from historical data. A case study using real misalignment data showed that the prediction results of the proposed calibration models improved significantly compared to existing calibration models. As an extension of linear calibration models, a nonlinear kernel calibration model was also presented. It turned out that linear and nonlinear calibration models of shaft misalignment conditions produced better prediction performance through the use of GA-based variable selection combined with OSC.     

Bifurcation and chaos analysis of nonlinear rotor system with axial-grooved gas-lubricated journal bearing support

Axial-grooved gas-lubricated journal bearings have been widely applied to precision instrument due to their high accuracy, low friction, low noise and high stability. The rotor system with axial-grooved gas-lubricated journal bearing support is a typical nonlinear dynamic system. The nonlinear analysis measures have to be adopted to analyze the behaviors of the axial-grooved gas-lubricated journal bearing-rotor nonlinear system as the linear analysis measures fail. The bifurcation and chaos of nonlinear rotor system with three axial-grooved gas-lubricated journal bearing support are investigated by nonlinear dynamics theory. A time-dependent mathematical model is established to describe the pressure distribution in the axial-grooved compressible gas-lubricated journal bearing. The time-dependent compressible gas-lubricated Reynolds equation is solved by the differential transformation method. The gyroscopic effect of the rotor supported by gas-lubricated journal bearing with three axial grooves is taken into consideration in the model of the system, and the dynamic equation of motion is calculated by the modified Wilson-0-based method. To analyze the unbalanced responses of the rotor system supported by finite length gas-lubricated journal bearings, such as bifurcation and chaos, the bifurcation diagram, the orbit diagram, the Poincar6 map, the time series and the frequency spectrum are employed. The numerical results reveal that the nonlinear gas film forces have a significant influence on the stability of rotor system and there are the rich nonlinear phenomena, such as the periodic, period-doubling, quasi-periodic, period-4 and chaotic motion, and so on. The proposed models and numerical results can provide a theoretical direction to the design of axial-grooved gas-lubricated journal bearing-rotor system.     

基于GRU-XGBoost的风电场功率短期预测

随着风电技术的不断发展,更多的风电机组并入电网运行。考虑到电网的安全性与稳定性,精确的风电场发电短期预测技术越发重要。在利用自适应噪声的完备经验模态分解(CEEMDAN)风电原始序列信号的基础上,采用GRU-XGBoost模型对非线性、非平稳的功率序列进行建模和预测,以提高模型的预测能力和泛化能力。首先利用CEEMDAN将风电功率原始序列分解为一系列不同时间尺度的分量,将分解后的信号输入GRU神经网络输出预测信号,再输入XGBoost进行校正。通过与多种预测模型进行比较证明此模型拥有更好的预测精度。     

Comparative analysis of neural network and regression based condition monitoring approaches for wind turbine fault detection

This paper presents the research results of a comparison of three different model based approaches for wind turbine fault detection in online SCADA data, by applying developed models to five real measured faults and anomalies. The regression based model as the simplest approach to build a normal behavior model is compared to two artificial neural network based approaches, which are a full signal reconstruction and an autoregressive normal behavior model. Based on a real time series containing two generator bearing damages the capabilities of identifying the incipient fault prior to the actual failure are investigated. The period after the first bearing damage is used to develop the three normal behavior models. The developed or trained models are used to investigate how the second damage manifests in the prediction error. Furthermore the full signal reconstruction and the autoregressive approach are applied to further real time series containing gearbox bearing damages and stator temperature anomalies.The comparison revealed all three models being capable of detecting incipient faults. However, they differ in the effort required for model development and the remaining operational time after first indication of damage. The general nonlinear neural network approaches outperform the regression model. The remaining seasonality in the regression model prediction error makes it difficult to detect abnormality and leads to increased alarm levels and thus a shorter remaining operational period. For the bearing damages and the stator anomalies under investigation the full signal reconstruction neural network gave the best fault visibility and thus led to the highest confidence level.     

基于组合模型的雷达故障预测分析

基于灰色 GM(1,1)模型和回归模型的建模原理,将两种模型进行拟合,建立了新的组合模型,并采用该模型对武器装备的故障进行预测。该组合模型充分利用了两种预测方法提供的信息,实现了两种模型之间功能和优势的互补,有效提高了预测精度。最后,以某型雷达发射机的等时距测量的输出电压为例估计系统的故障时间,并依此推断出该系统的故障发生时间。     

Brake squeal: Linear and nonlinear numerical approaches

“Brake squeal” groups a large set of high-frequency sound emissions from brake systems. They are generated during the braking phase and are characterized by a harmonic spectrum. The onset of squeal is due to an unstable behaviour occurring in linear conditions during the braking phase, and a general approach used by several authors to determine the system instabilities is the complex eigenvalues analysis. When the brake begins to squeal, the response of the system reaches a new limit cycle where the linear models cannot be used anymore. This paper presents the integration of two different numerical procedures to identify the mechanism bringing to squeal instability and to analyse its dynamics. The first approach is a finite element modal analysis of the brake system and is used to identify its eigenvalues and to relate them to the squeal occurrence. The second one is a specific finite element programme, Plast3, appropriate for nonlinear dynamic analyses in the time domain and is particularly addressed to study contact problems with friction between deformable bodies. This programme computes the contact stresses and permits to determine the dynamics of the system along the contact surface, both in the linear and nonlinear fields. The two models are compared and the onset of squeal is predicted both in the frequency domain by the linear model and in the time domain by the nonlinear one. The instability predictions, obtained by the two models, are discussed. To simplify the dynamics of its components, the study is carried out on a simple model, made of a disc, a small friction pad and a beam supporting the pad. The geometry of the model is related to an experimental set-up used to validate the models and to compare the numerical results with the experiments.     

基于AR_CLSTM的多元时间序列预测分析

时间序列是一种广泛应用于电量预测、汇率预测、太阳能发电量预测等各种领域的数据,预测其变化具有重要的意义.与LSTM相结合的编码器-解码器被广泛应用于多元时间序列预测.由于编码器只能将信息编码成固定长度的向量,因此模型的性能随着输入序列或输出序列长度的增加而迅速下降.为此,提出了基于编解码结构与线性回归的组合模型(AR CLSTM),该模型使用基于时间步的注意力机制使解码器能够自适应选择过去的隐藏状态并提取有用的信息,并利用卷积的结构学习多元时间序列不同维度之间的内在联系,同时结合了传统的线性自回归方法来学习时间序列的线性关系,从而实现在编解码结构上进一步降低时间序列预测的误差,改善多元时间序列的预测效果.实验结果表明,AR_CLSTM模型在不同的时间序列预测上表现良好,其均方根误差、均方误差、平均绝对误差均下降显著.     

Prediction of the machined surface quality of ball-end milling of H13 die steel using MLBP method

Abstract

Based on the characteristics of the surface quality prediction system of high-speed milling, the prediction model is used to predict the surface quality of analyzing the advantages of the two methods of using the multilinear and BP neural network model (MLBP) method. This article through the in-depth study of the surface quality, study the surface quality prediction based on the characteristics of multiinput multioutput nonlinear systems, respectively, established a linear regression equation, BP neural network model, and the surface quality of specific conditions to start prediction. The prediction results show that these prediction methods can play a special role as certain conditions. However, owing to the limitations of multiple linear regression and BP neural networks, their generalization ability and robustness cannot meet actual needs. Drawing on the idea of interpolation, and analyzing the advantages and disadvantages of linear regression and BP neural network to solve nonlinear problems, a new prediction method is developed. The main idea are to use interpolation method to insert preprediction under the premise of linear prediction; to process the values and obtain a unified prediction result from linear regression; to combine the experimental results from the pretreatment results; to use these input information as the input content of the BP neural network; to establish a training model based on the BP neural network model self-learning process. This training model predicts the quality of the machined surface. This method is abbreviated as the MLBP method. The experimental results and comparison of model prediction results show that this method can effectively improve the generalization ability and robustness of the prediction model, and further improve the model’s prediction accuracy.     

Method for identifying models of nonlinear systems using linear time periodic approximations

This work presents a new method for identifying models of nonlinear systems from experimental measurements. The system is first forced to oscillate in stable periodic orbit, and then a small impulsive disturbance force is used to perturb the system slightly from that orbit. One then measures the response until the system returns to the periodic orbit. If the nonlinearities in the system are sufficiently smooth and the perturbation from the periodic orbit is sufficiently small, then one can linearize the perturbed response about the periodic orbit and approximate the system as linear time periodic. One of a variety of methods can then be used to extract the time varying modal model of the system from the response. The extracted modes can be used to construct a time periodic state transition matrix and state coefficient matrix, which describe the system's nonlinear dynamics over a range of the states. The resulting model for the nonlinear system encompasses that portion of the state space is traversed by the system during its periodic orbit.     

Predicting engine reliability by support vector machines

Capturing the trends of engine failure data and predicting system reliability are very essential issues in engine manufacturing. The support vector machines (SVMs) have been successfully applied in solving nonlinear regression and times series problems. However, the application of SVMs to reliability forecasting is not widely explored. Therefore, to aim at examining the feasibility of SVMs in reliability predicting, this study is a first attempt to apply a SVM model to predict engine reliability. In addition, three other time series forecasting approaches, namely the Duane model, the autoregressive integrated moving average (ARIMA) time series model and general regression neural networks (GRNN), are used to compare the predicting performance. The experimental results show that the SVM model is a valid and promising alternative in reliability prediction.