Because reliable data-driven modeling methods require quantifying model uncertainty, an estimation of the parameter uncertainty is important. Usually maximum-likelihood methods with a known probability distribution of the noise are used. In practice, for many problems the stochastic properties of the errors are unknown and cannot be determined. In such cases, bounded error parameter estimation methods can be beneficial. These assume that the error lies within prior specified bounds. However, parameter estimation in the bounded error setting can be difficult and computationally expensive for data-driven modeling as a high number of parameters results. In this paper a new method for an approximate estimation with reduced effort of the set of feasible parameters for nonlinear data-driven modeling with Takagi-Sugeno fuzzy models is introduced. This permits to tackle large real-world problems. For this, a sampling based ray shooting method is proposed that guarantees an inner approximation of the feasible parameter set. The capability of the proposed method is demonstrated in two case studies, including one with data from an industrial hard turning process.

Recurrent Neural Networks (RNNs) are an internal dynamics approach to identify models from time series data. They have been successfully applied, e.g. in natural language, speech and video processing [1] and the identification of nonlinear state space models [2]. Conventional RNNs, such as the Elman-RNN, are notoriously hard to optimize, since they are highly initialization dependent, prone to slow convergence, and tend to converge to poor local minima. In recent years, the vanishing/exploding gradient phenomenon, which arises when employing gradient-based optimization techniques such as Backpropagation Through Time (BPTT), has been identified as the root cause of these difficulties. This led to the development of several new RNN-architectures, such as the Long Short-Term Memory (LSTM) and the Gated Recurrent Unit (GRU), which were intended to prevent the vanishing-gradient problem and surpassed conventional RNNs in all areas of application. However, it has been shown that the gradient also vanishes in Gated Units [3] and there is no work showing, that the rate of decay is lower than in Elman-RNN. This suggests that the underlying mechanisms responsible for their success are at least in part not yet fully understood. The purpose of this paper is to provide an alternative explanation for the superior performance of Gated Units by viewing them as nonlinear dynamical systems and studying the stability of their fixed points. This work expands on the work of Doya et al. [4] and Pascanu et al. [5], who studied the effects of bifurcation boundaries in the parameter space of Elman-RNNs with one internal state on gradient-based learning.

Gas leak inspection and gas leak quantification are nowadays of high relevance within the oil and gas industry as well as in many other industrial sectors. This has been driven by safety-related issues, economic losses and the considerable climate impact caused by such unwanted gas releases. Due to the latter, the efforts for developing new and more reliable measurement techniques for detecting and quantifying greenhouse gases such as methane have increased in the recent years. In this work, a stereo camera system based on optical gas imaging cameras is used for computing dense 3D velocity information, i.e. scene flow, of escaping gas structures. Here, the optical flow, the disparity and the disparity change in likely gas image regions are computed utilizing classical variational methods. The accuracy of the applied methods and their applicability under real conditions in a biogas plant are characterized and tested. The results show that the recovered 3D gas velocity field per camera frame approaches the average 3D velocity field of the measured gas structure. The accuracy of the used method is affected, among others, when the imaged gas structures exhibit a low contrast.

With the development of advanced cutting tool materials, hard turning has become a beneficial alternative to grinding in finishing of highly stressed mechanical components. However, a poor selection of cutting parameters may cause deterioration of the component's surface integrity and thus its fatigue life. Up to now, the control of surface integrity parameters is not possible as most of these quantities can not be measured during operation. Hence, suitable models are necessary to predict, e.g., the surface hardness and residual stresses induced by hard turning in-process. In order to achieve this, multiple regression and Takagi-Sugeno modeling are applied and compared in this contribution regarding their capability to predict axial and residual stress depth profiles depending on the cutting parameters and the initial hardness of the workpieces. Data from specimen machined on an industrial lathe and measured by X-ray diffraction are used for the case study.

Optimal experiment design (OED) is a well-developed concept for linear regression and linear dynamical modeling problems. In case of nonlinear models, the dilemma is that in order to evaluate the Fisher Information Matrix (FIM) for experiment design, the parameters to be estimated are required to evaluate the FIM. In case of locally affine Takagi-Sugeno (TS) models and D-optimal designs, even a `robust' sequential OED may not be sufficiently robust against wrong assumptions on partition parameters. As remedy, a two stage experiment design is proposed: It uses a space-filling design to estimate good initial TS model parameters. These are used to initialize a robust sequential FIM-based OED. The method is demonstrated for a nonlinear regression problem.

”Complex networks” is the term for a research area where complex systems are modeled by a graph to analyze their structural behavior. They are mostly used in the areas of social sciences, biology and physics. For example, complex networks are a proper method to describe and analyze the non-trivial characteristics depending on the interconnection in a society or between human organs. In the context of computational intelligence, this paper introduces an idea to transfer the methods of the area of complex networks to technical systems, and, fur-thermore, enhance them to permit analyzing dynamic behavior. As a basis for the method transfer a transfer-function-based graph is presented which allows modeling technical systems in the same way as complex networks. The potential to detect dynamical behavior, in addition to structural behavior, is demonstrated by a new algorithm that detects inverse response in interconnected systems based on methods of complex networks. The introduced algorithm provides a qualitative answer if inverse response behavior is possible between a pair of input and output of a system. Finally, two case studies are used to demonstrate the algorithm.

In order to automate routine inspections in large industrial environments, a mobile robotic system is being developed in the RoboGasInspector project. The robot's sensor-head consists of different instruments for remote gas sensing using multiple measurement principles. This contribution focuses on Tunable Diode Laser Absorption Spectroscopy (TDLAS). The measurement device allows quantitative gas concentration measurements and achieves high sensitivity due to the active measurement principle. On the contrary, TDLAS systems measure the concentration along a path at a time and require a diffusely reflecting background. Using a pan-tilt unit, objects and areas can be scanned. This paper describes the triangulation-maximization leak-localizing strategy (Tri-Max) used in the project RoboGasInspector. Results based on outdoor experiments in two industrial plants and a landfill dump are presented.

Models are commonly derived and their performance is assessed wrt. minimal prediction error on a closed data set. However, if no perfect model can be used, the degrees of freedom in modeling should be used to adjust the model to application-specific metrics. For model-based controller design, control-oriented performance metrics (e.g. performance wrt. to control-critical properties) are important, but not primarily prediction (i.e. prognosis- and simulation-oriented) ones. This motivates the derivation of control-specific models. The contribution introduces structured and quantitative measures on “model suitability for control” for the class of affine dynamic Takagi-Sugeno models. A method is suggested that derives control-specific dynamic models from a physical model given as a set of nonlinear differential equations. Within a case study, the proposed method demonstrates its significance: Using control-specific models improves control performance metrics such as set-point tracking quality, stability region and energy efficiency. Nonlinear dynamic modeling, Takagi-Sugeno systems, modeling for control performance metrics such as set-point tracking quality, stability region and energy efficiency.

Mobile robots for carrying out service tasks in industrial plants and on infrastructure gain importance e.g. due to aging assets and demographics. In order to identify key areas requiring research and development activities a questionnaire-based online survey was conducted in 2009. Addressed were participants working in industry knowing best about needs potentials obstacles and pitfalls of applying mobile service robots in their companies/industrial sector. This paper presents the surveying methods and the obtained results which mainly cover chemical and automotive industries due to their world-wide economic significance as well as their research and innovation intensity.

While other robots use in-situ measurements for gas leak detection and localization, we propose to apply remote sensing. It is easier and safer to conduct, permits rapid scans and is applicable to leak sources high up. An IR-optical sensor is used, exploiting spectral absorption effects of gases. Tailored leak detection and localization strategies are proposed. A simulation environment with a 3D model of the gas concentration field is used for developing and testing the detection and localization strategies. The system performance is demonstrated in a case study with a chemical plant.

Today, pressured air and gas leaks are typically detected using in-situ sensor technology. In this contribution the application of passive IR-thermography is proposed to permit remote leak detection by assessing the resulting temperature profile disturbance due to expansion of pressured gas. Remote measurements are advantageous as they are easier and safer to conduct. Scanning high-rise objects can be achieved using a ground-bound system without requiring complex climbing. In the paper a novel method for automated leak detection by feature extraction and pattern recognition is presented. This enables autonomous mobile robots with remote leak detection capability.

The performance of model-based controller design relies heavily on the quality and suitability of the utilized process model. This contribution proposes a fuzzy network based nonlinear controller design methodology. Fuzzy networks are a model approach combining high approximation quality with high interpretability. The input/output (I/O) models commonly used for identification are transformed to fuzzy state-space models. Transferring and adjusting methods from linear state-space theory a control concept consisting of a fuzzy state controller and an adaptive set-point filter for nonlinear dynamic processes is deduced. The capability of the method is demonstrated for a hydraulic drive.