Detection of abrupt change and applications in sensor decalibration monitoring

The primary objective of fault detection is to detect abrupt undesirable changes in a process at an early stage. This early detection has a potential of preventing loss of production and equipment damage due to these undesirable changes, thus reducing process downtime. This paper details the implementation of some parametric fault detection techniques for sensor decalibration monitoring. A parametric fault detection approach that is handled in depth in this paper is the local approach. This approach developed by Benveniste, Basseville, and Moustakides [Benveniste, A., Basseville, M., and Moustakides, G., The asymptotic local approach to change detection and model validation. IEEE Trans. Autom. Control AC-32 (7), 583-592 (1987)] offers a computationally inexpensive way to attain the objective of monitoring changes in model parameters. However, the algorithm in its original formulation is not applicable to certain processes such as sensors. Therefore, the local approach is coupled with other estimation algorithms such as the input independent Kalman filter to derive a robust sensor decalibration monitoring algorithm. The proposed fault detection algorithm is applied to a pilot scale process for evaluation of its performance.     

二阶线性系统调参共振的特征信号检测

以二阶线性系统为信息检测处理模型,提出二阶线性系统调参共振的特征信号检测方法。该方法利用线性系统的共振特性,得到系统响应最大值随固有频率变化的特性曲线,根据曲线中的极大值即可识别噪声中的特征信号。仿真和转子故障诊断实验表明,所提出的方法原理简单,能为实际信号检测与处理提供一种可行实用的方法。     

一种基于电压和频率的金属探测方法

首先在能耗式检测电路基础上，提出一种利用检测线圈中电压和频率的变化来探测金属的原理，继而分析了不同属性金属和物料对检测电路中电压和频率变化的影响及这种变化的规律，最后，提出利用这种变化规律在不同应用中的检测方案，实验结果表明了理论分析的正确性和实用性。     

AR模型和分形几何在设备状态监测中的应用研究

状态特征指标对机械的状态监测和故障诊断具有重要意义。本文提出应用机械设备工作状态下噪声信号自回归模型的关联维数来描述设备在不同工作状态下的特征,进而实现对状态的监测、识别和分类。文中通过实验证明,设备在相同工作状态下,噪声信号的AR模型参数具有相近的关联维数,在不相同状态下则有明显不同的关联维数。因此关联维数不仅可以作为状态监测与识别和分类的重要依据,而且可以作为其他特征提取方法的补充。此方法对设备状态监测准确率的提高具有明显的作用。     

Sparse Autoencoder-based Multi-head Deep Neural Networks for Machinery Fault Diagnostics with Detection of Novelties

Supervised fault diagnosis typically assumes that all the types of machinery failures are known.However,in practice unknown types of defect,i.e.,novelties,may occur,whose detection is a challenging task.In this paper,a novel fault diagnostic method is developed for both diagnostics and detection of novelties.To this end,a sparse autoencoder-based multi-head Deep Neural Network(DNN)is presented to jointly learn a shared encoding representation for both unsupervised reconstruction and supervised classification of the monitoring data.The detection of novelties is based on the reconstruction error.Moreover,the computational burden is reduced by directly training the multi-head DNN with rectified linear unit activation function,instead of performing the pre-training and fine-tuning phases required for classical DNNs.The addressed method is applied to a benchmark bearing case study and to experimental data acquired from a delta 3D printer.The results show that its performance is satisfactory both in detection of novelties and fault diagnosis,outperforming other state-of-the-art methods.This research proposes a novel fault diagnostics method which can not only diagnose the known type of defect,but also detect unknown types of defects.     

An improved spectral analysis of the stray flux component for the detection of air-gap irregularities in squirrel cage motors

For machines? monitoring purpose, the classical motor current signature analysis has shown its weakness in distinguishing the eccentricity occurrence in presence of others mechanical faults. Although Park?s vector approach can cover this drawback, the high cost due to the requirement to use three current sensors associated with an advanced processing technique, makes it less desired by industrialists. In this paper, we suggest an alternative diagnosis method based on a suitable processing of the stray flux data. The experimental results have revealed the potential of a simple search coil for the detection and the distinction of the accurate eccentricity nature even in presence of similar mechanical faults.     

Deep particle image velocimetry supervised learning under light conditions

Particle image velocimetry (PIV) is an important fluid visualization technology which extracts the velocity field from two successive particle images. Recently, some researchers have begun to use convolutional neural network (CNN) to tackle the PIV problem successfully. Some supervised learning methods make use of the PIV dataset with ground truth for network training. However, the existing dataset is composed of pairs of particle images under ideal light conditions and does not take into account the changes in actual experimental conditions. In this paper, we firstly generated a new and more challenging dataset called Light-PIV which fully simulates the change of the brightness of particle images in the real PIV experiment. Secondly, we present here a novel approach for fluid motion estimation which is based on an optical flow network LiteFlowNet. The proposed approach is verified by the application to a diversity of synthetic and experimental PIV images. We not only improve the structure, but also combine the traditional prior assumptions knowledge with the loss function to better guide the network training. The proposed approach is verified by the application to a diversity of synthetic and experimental PIV images. The experimental results show that our proposed method has advantages of high accuracy, obtaining detailed information and strong robustness in our PIV dataset compared with classical PIV methods such as HS optical flow and WIDIM, and even outperforms these existing approaches in some flow cases.     

多尺度特征自适应融合的气动控制阀故障诊断

气动控制阀作为过程工业典型的终端执行机构,由于故障发生率高、故障类型繁多,导致故障识别难度大,且故障后果 严重,因此对其进行智能的故障检测和诊断具有重要的实际意义。 本文提出了一种多尺度特征自适应融合网络用于气动控制 阀故障诊断。 首先,搭建了融合自注意力机制的多尺度特征提取网络自动提取信号的空间特征和细节特征。 然后,设计了权重 自适应特征融合网络对多尺度特征加权融合,提高模型对故障特征的表征能力。 最后,由长短时记忆神经网络和 SoftMax 函数 实现特征识别和故障分类。 实验结果表明,该模型在 DAMADICS 阀门基准实验平台上的平均检测准确率达到 96. 82% ,均高于 其他对比模型。 与最近发表文献中的检测结果对比发现,本文开发的模型在可检测的故障数量和检测准确率方面也具有一定 的优势,并且通过实验验证了模型的检测性能。     

Experimental investigation of heat conduction performance and development of automatic temperature measurement device on modular artillery charge system

The heat conduction performance experiment is conducted on the modular artillery charge system (MACS). Utilizing the experimental measurement system, the change history of the modular charge temperature is obtained. On the basis of the heat conduction performance experiment of modular propellant charge, an unsteady-state heat conduction model describing the temperature change of the MACS is built and the finite-difference implicit schemes are theoretically deduced using the volume equilibrium method for numerical simulation. The validation of the numerical model is checked through compared with the experimental results. An automatic online temperature measurement device on the MACS is developed, based on the non-contact measurement method which is proposed in this paper. As a part of the device, an initial charge temperature sensor (ICTS) is also developed according to the similarity principle. The temperature measurement device where the numerical codes are embedded is assembled in the turret to calculation successively the temperature change of the modular charge with the environment temperature of ammunition rack. Meanwhile, for trajectory calculation and firing data correction, the temperature information obtained from the device may be simultaneously transferred to the gunner task terminal computer through a Controller Area Network (CAN) bus interface.     

An experimental demonstration of tailored excitations for improved damage detection in the presence of operational variability

We have shown in previous work that the parameters governing the dynamic evolution of a system of ordinary differential equations may be modified via an evolutionary algorithm to yield excitations that improve damage detection sensitivity in a computational model of a structural health monitoring application. In this work we use the same method to develop improved excitations for an experimental system. Improvement in change detection sensitivity is shown for several generated excitations. In addition, we present an excitation that increases the robustness of the damage detection feature to the types of parameter perturbations that arise not from damage, but from environmental sources such as temperature change.     

Implementation of a new fuzzy vector control of induction motor

The aim of this paper is to present a new approach to control an induction motor using type-1 fuzzy logic. The induction motor has a nonlinear model, uncertain and strongly coupled. The vector control technique, which is based on the inverse model of the induction motors, solves the coupling problem. Unfortunately, in practice this is not checked because of model uncertainties. Indeed, the presence of the uncertainties led us to use human expertise such as the fuzzy logic techniques. In order to maintain the decoupling and to overcome the problem of the sensitivity to the parametric variations, the field-oriented control is replaced by a new block control. The simulation results show that the both control schemes provide in their basic configuration, comparable performances regarding the decoupling. However, the fuzzy vector control provides the insensitivity to the parametric variations compared to the classical one. The fuzzy vector control scheme is successfully implemented in real-time using a digital signal processor board dSPACE 1104. The efficiency of this technique is verified as well as experimentally at different dynamic operating conditions such as sudden loads change, parameter variations, speed changes, etc. The fuzzy vector control is found to be a best control for application in an induction motor.     

Damage identification by multi-model updating in the modal and in the time domain

Computational model updating techniques are used to adjust selected parameters of finite element models in order to make the models compatible with experimental data. This is done by minimizing the differences (residuals) of analytical and experimental data, for example, natural frequencies and mode shapes by numerical optimization procedures. For a long-time updating techniques have also been investigated with regard to their ability to localize and quantify structural damage. The success of such an approach is mainly governed by the quality of the damage model and its ability to describe the structural property changes due to damage in a physical meaningful way. Our experience has shown that due to unavoidable modelling simplifications and measurement errors the changes of the corresponding damage parameters do not always indicate structural modifications introduced by damage alone but indicate also the existence of other modelling uncertainties which may be distributed all over the structure. This means that there are two types of parameters which have to be distinguished: the damage parameters and the other parameters accounting for general modelling and test data uncertainties. Although these general parameters may be physically meaningless they are necessary to achieve a good fit of the test data and it might happen that they cannot be distinguished from the damage parameters. For complex industrial structures it is seldom possible to generate unique structural models covering all possible damage scenarios so that one has to expect, that the parameters introduced for describing the damage will not be fully consistent with the physical reality. Even then the change of such parameters identified from test data taken continuously or temporarily over the time may serve as a feature for structural health monitoring. It is well known that low-frequency modal test data or static response data are not very well suited for detecting and quantifying localized small size damage. Time domain response data from impact tests carry high-frequency information which usually is lost when experimental modal data are utilized for damage identification. Even so only little literature was found addressing the utilization of experimental time histories for model updating in conjunction with damage identification.In the present paper we summarize the methodology of computational model updating and report about our experience with damage identification using two different model updating techniques. The first is based on classical modal residuals (natural frequencies and mode shapes) which is extended to allow for simultaneous updating of two models, one for the initial undamaged structure and the second for the damaged structure using the test data of both states (multi-model updating). The second technique uses residuals composed of measured and analytical time histories. Time histories have the advantage of carrying high-frequency information which is beneficial for the detection of local damage and which usually is lost when modal residuals are used. Both techniques have been applied to the same beam structure consisting of two thin face sheets which were bonded together by an adhesive layer. It was the aim of this application to study the performance of the two techniques to localize and quantify the damage which was introduced locally in the adhesive layer.     

免疫原理在设备故障诊断中的应用原理及模型

提出了一种基于人工免疫的故障诊断进化学习模型及其相应的算法.介绍了该诊断模型的理论基础,描述了诊断的步骤方法,并应用于机床齿轮箱故障检测和诊断问题.实验结果表明了所提出方法的有效性.     

New thermal contrast definition for defect characterization by active thermography

Active thermography is a relatively young method in comparison to the classical, standardized methods of non-destructive testing. Many papers present a possibility of using this method for detection of some types of defects such as delamination, cracks or skin holes. It is important to determine the limitation of this method and possible areas of application. This paper presents an experimental setup for defect detection and characterization lying in the subsurface layers of the tested material by means of an active thermography. The work proposes two versions of a new kind of thermal contrast, called “filtered contrast – FC” and “relative filtered contrast – RFC”, which allow to detect and to characterize defects under some assumptions on the physical and thermal parameters of the tested material and defect. In opposition to the known definitions of the thermal contrast, neither a reference thermogram nor a defect-free area is necessary. Additionally, this contrast is less sensitive to the non-uniformity of the heat disposal over the material surface than classical contrasts. Theoretical considerations are illustrated by experimental results. It has to be underlined that it is a preliminary study for simple two-layer structure.     

A method for detection and diagnosis on batch fermentations

In this work we present some basic ideas about detection and diagnosis of faults and abrupt dynamic changes in batch fermentations. Our work focuses on the simultaneous use of two detection methods (residual based and balances based) within the estimation procedure. The idea behind the use of both methods is that the weakness of one of them can be compensated by the use of the other one. Thus the simultaneous use of both methods allows detecting and possibly isolating a wide range of faults. Observations such as the effect of nonlinearities on the detection tests and robustness to model uncertainty are discussed. Numerical results on a particular case, the xanthan gum batch fermentation, are presented. Simulated faults and abnormal behaviors were promptly detected but diagnostics showed mixed results.     

Fusion of qualitative bond graph and genetic algorithms: a fault diagnosis application

In this paper, the problem of fault diagnosis via integration of genetic algorithms (GA's) and qualitative bond graphs (QBG's) is addressed. We suggest that GA's can be used to search for possible fault components among a system of qualitative equations. The QBG is adopted as the modeling scheme to generate a set of qualitative equations. The qualitative bond graph provides a unified approach for modeling engineering systems, in particular, mechatronic systems. In order to demonstrate the performance of the proposed algorithm, we have tested the proposed algorithm on an in-house designed and built floating disc experimental setup. Results from fault diagnosis in the floating disc system are presented and discussed. Additional measurements will be required to localize the fault when more than one fault candidate is inferred. Fault diagnosis is activated by a fault detection mechanism when a discrepancy between measured abnormal behavior and predicted system behavior is observed. The fault detection mechanism is not presented here.     

Distance similarity matrix using ensemble of dimensional data reduction techniques: Vibration and aerocoustic case studies

Vibration and acoustic-based health-monitoring techniques are used in the literature to monitor structural health under dynamic environment. In this paper, we propose a damage detection and monitoring method based on a distance similarity matrix of dimensionally reduced data wherein redundancy therein is removed. The matrix similarity approach is generic in nature and has the capability of multiscale representation of datasets. To extract damage-sensitive features, dimensional reduction techniques are applied and compared. An ensemble method of dimensional reduction feature outputs is presented and applied to two case studies. The results supports why ensembles can often perform better than any single-feature extraction method. For the first case study, aeroacoustic datasets are collected from controlled scaled experimental tests of controlled known damaged subscale wing structure. For the second case study, a vibration experiment study is used for abrupt change detection and tracking. The results of the two case studies demonstrate that the proposed method is very effective in detecting abrupt changes and the ensemble method developed here can be used for deterioration tracking.     

Adaptive Change Detection for Long-Term Machinery Monitoring Using Incremental Sliding-Window

Detection of structural changes from an operational process is a major goal in machine condition monitoring. Existing methods for this purpose are mainly based on retrospective analysis, resulting in a large detection delay that limits their usages in real applications. This paper presents a new adaptive real-time change detection algorithm, an extension of the recent research by combining with an incremental sliding-window strategy, to handle the multi-change detection in long-term monitoring of machine operations. In particular, in the framework, Hilbert space embedding of distribution is used to map the original data into the Re-producing Kernel Hilbert Space (RKHS) for change detection; then, a new adaptive threshold strategy can be developed when making change decision, in which a global factor (used to control the coarse-to-fine level of detection) is introduced to replace the fixed value of threshold. Through experiments on a range of real testing data which was collected from an experimental rotating machinery system, the excellent detection performances of the algorithm for engineering applications were demonstrated. Compared with state-of-the-art methods, the proposed algorithm can be more suitable for long-term machinery condition monitoring without any manual re-calibration, thus is promising in modern industries.     

Comparison of system performance on hot-gas bypass and variable speed compressor in an oil cooler for machine tools

This paper compares the system performance for two representative capacity control schemes, hot-gas bypass and variable speed compressor in an oil cooler for machine tools. An empirical linear first-order model of each controlled system was obtained from experiments by imposing a stepwise control signal to each actuator of the system. General proportional-integral controllers are designed based on the empirical transfer function models in order to control the target temperature of the oil cooler system. The experiments of starting and thermal load change were conducted to compare their control performance with each other. Especially, coefficient of performance (COP) of the two control schemes under the partial load state was also analyzed in detail. From the analyses of experimental results, the control performance of the target temperature in the two control schemes had almost the same control accuracy of ±0.1 °C at steady state. However, the COP of a variable speed compressor was as many as five-times greater than that of hot-gas bypass in comparable minimum partial load state of 0.5 kW.     

Virtual prototype and experimental research on gear multi-fault diagnosis using wavelet-autoregressive model and principal component analysis method

Gear systems are an essential element widely used in a variety of industrial applications. Since approximately 80% of the breakdowns in transmission machinery are caused by gear failure, the efficiency of early fault detection and accurate fault diagnosis are therefore critical to normal machinery operations. Reviewed literature indicates that only limited research has considered the gear multi-fault diagnosis, especially for single, coupled distributed and localized faults. Through virtual prototype simulation analysis and experimental study, a novel method for gear multi-fault diagnosis has been presented in this paper. This new method was developed based on the integration of Wavelet transform (WT) technique, Autoregressive (AR) model and Principal Component Analysis (PCA) for fault detection. The WT method was used in the study as the de-noising technique for processing raw vibration signals. Compared with the noise removing method based on the time synchronous average (TSA), the WT technique can be performed directly on the raw vibration signals without the need to calculate any ensemble average of the tested gear vibration signals. More importantly, the WT can deal with coupled faults of a gear pair in one operation while the TSA must be carried out several times for multiple fault detection. The analysis results of the virtual prototype simulation prove that the proposed method is a more time efficient and effective way to detect coupled fault than TSA, and the fault classification rate is superior to the TSA based approaches. In the experimental tests, the proposed method was compared with the Mahalanobis distance approach. However, the latter turns out to be inefficient for the gear multi-fault diagnosis. Its defect detection rate is below 60%, which is much less than that of the proposed method. Furthermore, the ability of the AR model to cope with localized as well as distributed gear faults is verified by both the virtual prototype simulation and experimental studies.