Search

Proton Exchange Membrane Fuel Cells are a favorite technology for decarbonizing the transportation sector. However, their large-scale democratization is hampered by their high cost compounded by their unsatisfactory lifespan. To anticipate potential degradation while keeping improving performance, it is essential to maintain an acceptable humidity range inside the cells, especially at the membrane level. However, membrane humidity level is not directly measurable, alternative techniques must be considered to recover this key variable. Here, we develop a real-time software sensor of the membrane water content at the fuel cell's heart. We build a model describing the membrane water balance, electrochemical behavior, and species mass balance. We then reduce the model and perform an Adaptive Extended Kalman Filter. We perform sensitivity analyses in both steady-state and transient conditions. We validate the filter on a "Worldwide Harmonized Light Vehicles Test Cycles" test procedure. Finally, we obtain a fast and accurate model-based software sensor. • Real-time capable software sensor for PEMFC membrane water content. • Observer based on membrane model accounting for most of major transport phenomena. • Adaptive Extended Kalman filter applied to differential-algebraic equations. • Sensitivity analysis of membrane water content shows a 5% coefficient of variation.

Deep drawing is one of the most important forming processes for the forming of flat sheet blanks, where the formation of wrinkles and the appearance of cracks can be a problem, especially in areas of high geometric complexity. The local increase of the temperature in these critical areas can help to improve the formability of the material and thus reduce defects. The present paper aims at a targeted temperature control of the die of a deep-drawing mold. For this sensors are placed systematically to develop an estimator for the spatial–temporal temperature evolution to subsequently realize tracking control using the embedded actuation devices. A continuum representation of the temperature distribution in the die is derived and transferred to a high order finite element (FE) approximation to take the complex-shaped geometry of the tool into account. Parameter identification is performed based on measurement data to improve the accuracy of the FE approximation and model order reduction (MOR) techniques are applied to determine a sufficiently low order system representation. A mixed-integer optimization problem is formulated and solved making use of different formulations of the observability Gramian to determine the optimal sensor locations and a Kalman filter is designed as an estimator based on a reduced order model. Moreover, a linear-quadratic regulator with integral part combined with the Kalman filter is developed to react efficiently towards disturbances. Finally this theoretical framework is tested in a real experiment. • Model derivation and Finite Element approximation for a deep drawing tool with high geometric complexity. • Experimental parameter identification. • Comparison and evaluation of different model order reduction techniques. • Optimal sensor placement, estimator design, and temperature control design based on the reduced order model. • Experimental validation of the setup. [ABSTRACT FROM AUTHOR]

This paper addresses the design and evaluation of a robust observer for second order bioprocesses considering unknown bounded disturbance terms and uncertainty in the dynamics of the unknown and known states. The observer design and the stability analysis are based on dead-zone Lyapunov functions, and a detailed procedure is provided. The transient response bounds and the convergence region of the unknown observer error are determined in terms of the disturbance bounds, considering persistent but bounded disturbances in the dynamics of both the known and unknown observer errors. This is a significant contribution to closely related observer design studies, in which the transient response bounds are determined, but persistent and bounded disturbances are not considered in the dynamics of the known observer error. Other important contributions are: (i) the procedure for defining the observer parameters is significantly simpler than common observer designs, since a solution to the Ricatti equation, solution to LMI constraints, or the accomplishment of eigenvalue inequality conditions are not required; (ii) discontinuous signals are not used in the observer; and (iii) the effect of the gain sign associated with the unknown state in the dynamics of the known state is explicitly and clearly considered in the observer design and in the convergence study. In addition, the guidelines for selecting the observer parameters are provided. Numerical simulation confirms the stability analysis results: the observer errors converge within a short time, with a low estimation error, if observer-parameters are properly defined. [ABSTRACT FROM AUTHOR]

This paper addresses the design and evaluation of a robust observer for second order bioprocesses considering unknown bounded disturbance terms and uncertainty in the dynamics of the unknown and known states. The observer design and the stability analysis are based on dead-zone Lyapunov functions, and a detailed procedure is provided. The transient response bounds and the convergence region of the unknown observer error are determined in terms of the disturbance bounds, considering persistent but bounded disturbances in the dynamics of both the known and unknown observer errors. This is a significant contribution to closely related observer design studies, in which the transient response bounds are determined, but persistent and bounded disturbances are not considered in the dynamics of the known observer error. Other important contributions are: (i) the procedure for defining the observer parameters is significantly simpler than common observer designs, since a solution to the Ricatti equation, solution to LMI constraints, or the accomplishment of eigenvalue inequality conditions are not required; (ii) discontinuous signals are not used in the observer; and (iii) the effect of the gain sign associated with the unknown state in the dynamics of the known state is explicitly and clearly considered in the observer design and in the convergence study. In addition, the guidelines for selecting the observer parameters are provided. Numerical simulation confirms the stability analysis results: the observer errors converge within a short time, with a low estimation error, if observer-parameters are properly defined.

A new method for on-line dynamics monitoring of three physiological states of fed-batch E. coli fermentation is proposed. The method is based on the acetate kinetics accepted as adaptive key parameter for recognition of current physiological state. Software sensors of main kinetic parameters are derived. They are connected in cascade scheme with inputs on-line measurements of acetate and glucose concentrations. In the outputs the on-line information of three biomass growth rates and biomass concentration is received. Tuning of the estimation algorithms is realized. The efficiency of the proposed method is investigated by simulations using a new biochemical process model. Discussion about the accuracy of obtained estimates and their relationship with the values of tuning parameters is done. The proposed method could be applied for control algorithms design for each physiological state.

The paper presents tools to model low speed airflow coming from a turbulent machine. This low speed flow have instabilities who generate noise disturbances in the environment. The aim of the study proposed in this paper, is the using of cyclostationary tools with audio signals to model this airflow and detect the noisy frequencies to eliminate this noise. This paper also deals with the extraction in real time of the frequency corresponding to the noise nuisance. This extraction makes it possible to build a software sensor. This software sensor can be used to estimate the air flow rate and also to control a future actuator which will reduce the intensity of the noise nuisance. This paper focuses on the characteristic of the sound signal (property of cyclostationarity) and on the development of a software sensor. The results are established using an experimental setup representative of the physical phenomenon to be characterised.

Internal Circulation (IC) anaerobic systems are especially suitable when plant designs that involve both small footprints and high organic loading rates (>25 kg COD m −3 d −1 ) are required. However, given that operating anaerobic processes at high organic loads increases their vulnerability to external disturbances, real-time indicators of the stability conditions become particularly pertinent for IC reactors. This paper addresses the design and full-scale validation of a software sensor that uses only feeding flow-rate and biogas flow-rate measurements to classify the operating conditions of the reactor as “strongly”, “moderately” or “weakly” stable. A simulation-based procedure was used to design the software sensor and configure its parameters. Then, the performance of the software sensor was tested under real conditions in a full-scale IC reactor of 1679 m 3 installed in a recycled paper mill (RPM). [ABSTRACT FROM AUTHOR]

Microalgae are well known for their ability to accumulate lipids intracellularly, which can be used for biofuels and mitigate CO2 emissions. However, due to economic challenges, microalgae bioprocesses have maneuvered towards the simultaneous production of food, feed, fuel, and various high-value chemicals in a biorefinery concept. On-line and in-line monitoring of macromolecules such as lipids, proteins, carbohydrates, and high-value pigments will be more critical to maintain product quality and consistency for downstream processing in a biorefinery to maintain and valorize these markets. The main contribution of this review is to present current and prospective advances of on-line and in-line process analytical technology (PAT), with high-selectivity - the capability of monitoring several analytes simultaneously - in the interest of improving product quality, productivity, and process automation of a microalgal biorefinery. The high-selectivity PAT under consideration are mid-infrared (MIR), near-infrared (NIR), and Raman vibrational spectroscopies. The current review contains a critical assessment of these technologies in the context of recent advances in software and hardware in order to move microalgae production towards process automation through multivariate process control (MVPC) and software sensors trained on "big data". The paper will also include a comprehensive overview of off-line implementations of vibrational spectroscopy in microalgal research as it pertains to spectral interpretation and process automation to aid and motivate development. [ABSTRACT FROM AUTHOR]

In the context of motorcycle, we can assist to an increasing interest toward semi-active suspension control systems able to improve both the comfort and the passenger’s safety in both racing and original equipment manufacturer applications. Such systems implement suitable strategies based on the measure of several quantities, among which the relative velocity of the wheels respect to the vehicle body with the aim of regulating in real-time the damping forces. The actual effectiveness of such strategy strongly depends on the reliability and accuracy of the data measured by the sensors involved in the control loop. Due to their simplicity and good performance in terms of linearity, the most used sensors for suspension displacement measurements are based on linear potentiometers but such kind of sensors suffer of wear and tear and aging higher than the other sensors involved in the control loop strategy. As a consequence, the fault detection of such sensor is strongly recommended to avoid wrong and in some cases dangerous suspension behaviors. To this aim, in this paper a Fault Detection scheme for the rear suspension stroke sensor is designed and verified. The residual generation is based on the use of a Nonlinear Auto-Regressive with eXogenous inputs (NARX) network which is able to effectively take into account for the system nonlinearity. Experimental results have proven the good promptness and reliability of the scheme in detecting different kind of faults as “un-calibration faults” (e.g. due to slight variations of the input/output sensor curve), “hold-faults” (e.g. due to the breaking of the potentiometer cursor), “open circuit” and “short circuit” (e.g. due to electrical interruptions and short circuits, respectively). In addition, to verify the feasibility of a real-time implementation on actual processing units employed in such context, the scheme has been successfully implemented on a microcontroller STM32 based on the general-purpose ARM-M4 architecture. The validation tests and analysis have shown that the proposed Instrument Fault Detection scheme could be successfully developed on these kind of architectures by assuring a real-time operating. [ABSTRACT FROM AUTHOR]

Chemical systems are often characterized by a number of peculiar properties that create serious challenges to state estimator algorithms. They may include hard nonlinear dynamics, states subject to some constraints arising from a physical nature of the process (for example, all chemical concentrations must be nonnegative), and so on. The classical Extended Kalman Filter (EKF), which is considered to be the most popular state estimator in practice, is shown to be ineffective in chemical systems with infrequent measurements. In this paper, we discuss a recently designed version of the EKF method, which is grounded in a high-order Ordinary Differential Equation (ODE) solver with automatic global error control. The implemented global error control boosts the quality of state estimation in chemical engineering and allows this newly built version of the EKF to be an accurate and efficient state estimator in chemical systems with both short and long waiting times (i.e., with frequent and infrequent measurements). So chemical systems with variable sampling periods are algorithmically admitted and can be treated as well. [ABSTRACT FROM AUTHOR]

Automatisation and digitalisation of laboratory processes require adequate online measurement techniques. In this paper, we present affordable and simple means for non-invasive measurement of biomass concentrations during cultivation in shake flasks. Specifically, we investigate the following research questions. Can images of shake flasks and their content acquired with smartphone cameras be used to estimate biomass concentrations? Can machine vision be used to robustly determine the region of interest in the images such that the process can be automated? To answer these questions, 18 experiments were performed and more than 340 measurements taken. The relevant region in the images was selected automatically using K-means clustering. Statistical analysis shows high fidelity of the resulting model predictions of optical density values that were based on the information embedded in colour changes of the automatically selected region in the images.

In order to manage water resources, a software sensor model was designed to estimate water quality using a hybrid fuzzy neural network (FNN) in Guangzhou section of Pearl River, China. The software sensor system was composed of data storage module, fuzzy decision-making module, neural network module and fuzzy reasoning generator module. Fuzzy subtractive clustering was employed to capture the character of model, and optimize network architecture for enhancing network performance. The results indicate that, on basis of available on-line measured variables, the software sensor model can accurately predict water quality according to the relationship between chemical oxygen demand (COD) and dissolved oxygen (DO), pH and NH4+–N. Owing to its ability in recognizing time series patterns and non-linear characteristics, the software sensor-based FNN is obviously superior to the traditional neural network model, and its R (correlation coefficient), MAPE (mean absolute percentage error) and RMSE (root mean square error) are 0.8931, 10.9051 and 0.4634, respectively. [ABSTRACT FROM AUTHOR]

Semi-active suspension control needs to measure the relative velocity of the wheels respect to the vehicle body to regulate the damping forces. Linear potentiometers are the most used sensors in racing for linearity and simplicity, but they are expensive for mass market and unreliable in the long run. A solution to limit the cost of the vehicle electronics is the suitable exploitation of the analytical redundancy among the measured quantities of interest. The paper describes the application of a systematic approach to the optimum design of a soft displacement sensor, which may be adopted with the rear suspension of the two-wheeled vehicles. Focus is mainly devoted to the real-time exploitation of recurrent Artificial Neural Network within a typically adopted instrumentation set. Experimental results concerning with the soft sensor enable to develop an effective Instrument Fault Detection and Isolation scheme in order to improve the system reliability. [ABSTRACT FROM AUTHOR]

This work proposes an on-line hybrid intelligent control system based on a genetic algorithm (GA) evolving fuzzy wavelet neural network software sensor to control dissolved oxygen (DO) in an anaerobic/anoxic/oxic process for treating papermaking wastewater. With the self-learning and memory abilities of neural network, handling the uncertainty capacity of fuzzy logic, analyzing local detail superiority of wavelet transform and global search of GA, this proposed control system can extract the dynamic behavior and complex interrelationships between various operation variables. The results indicate that the reasonable forecasting and control performances were achieved with optimal DO, and the effluent quality was stable at and below the desired values in real time. Our proposed hybrid approach proved to be a robust and effective DO control tool, attaining not only adequate effluent quality but also minimizing the demand for energy, and is easily integrated into a global monitoring system for purposes of cost management. [ABSTRACT FROM AUTHOR]

A new method for on-line dynamics monitoring of three physiological states of fed-batch E. coli fermentation is proposed. The method is based on the acetate kinetics accepted as adaptive key parameter for recognition of current physiological state. Software sensors of main kinetic parameters are derived. They are connected in cascade scheme with inputs on-line measurements of acetate and glucose concentrations. In the outputs the on-line information of three biomass growth rates and biomass concentration is received. Tuning of the estimation algorithms is realized. The efficiency of the proposed method is investigated by simulations using a new biochemical process model. Discussion about the accuracy of obtained estimates and their relationship with the values of tuning parameters is done. The proposed method could be applied for control algorithms design for each physiological state. [ABSTRACT FROM AUTHOR]

This paper is concerned with the monitoring of a biodiesel production process, more specifically with the monitoring of the esterification of grease trap wastes, a low quality feedstock for biodiesel production typically characterized by its high content of Free Fatty Acids (FFAs). The esterification takes place in a jacketed Continuous Stirred Tank Reactor (CSTR). A reset observer is designed and applied in order to provide on-line estimation of the concentration of FFAs from temperature measurements within the CSTR. In addition, the proposed reset observer is compared to two other observers (classical fuzzy observer and extended Kalman filter). According to a multiple range test conducted for analyzing the estimation error, the monitoring task for the process under study has been better fulfilled by the reset observer which is able to update the estimation results every instant when the measurements were available. [ABSTRACT FROM AUTHOR]

In this paper, a robust software sensor with online estimation of stator resistance applied to a wind energy conversion system (WECS) using an induction machine (IM) was developed. The mechanical speed control requires a specific estimation of the flux which is dependent on the IM stator resistance. As a consequence, this estimation may face some problems because of the stator resistance variation. In this paper, a robust software sensor was designed and developed for online stator resistance estimation under parametric variations. The estimated flux and mechanical speed as well as stator resistance are given by the software sensor. The robustness and stability study of the suggested software sensor were proven and validated by simulation results. [ABSTRACT FROM AUTHOR]

The feeding of a fed-batch cultivation producing recombinant protein was controlled by a soft sensor set-up. It was assumed that the control approach could be based on the cell's production of carbon dioxide and that this parameter indicates the metabolic state occurring at induced protein expression. The soft sensor used the on-line signals from a carbon dioxide analyser and a near-infrared (NIR) probe for biomass to estimate the specific production rate ( q CO 2 ) . Control experiments were carried out with various q CO 2 set-points where we observe that the feeding of nutrients to the culture could easily be controlled and resulted in a decreased variability compared to uncontrolled cultivations. We therefore suggest that this control approach could serve as an alternative to other commonly applied methods such as controlling the cell's overflow metabolism of acetate or the cell's specific growth rate. However, further studies of the internal metabolic fluxes of CO 2 during protein expression would be recommended for a more precise characterization of the relationship between q CO 2 and protein expression in order to fully interpret the control behaviour. [ABSTRACT FROM AUTHOR]

This paper focuses on the specific growth rate estimation problem in a Polyhydroxybutyrate bioplastic production process by industrial fermentation. The kinetics of the process are unknown and there are uncertainties in the model parameters and inputs. During the first hours of the growth phase of the process, biomass concentration can be measured online by an optical density sensor, but as cell density increases this method becomes ineffective and biomass measurement is lost. An asymptotic observer is developed to estimate the growth rate for the case without biomass measurement based on corrections made by a pH control loop. Furthermore, an exponential observer based on the biomass measurement is developed to estimate the growth rate during the first hours, which gives the initial condition to the asymptotic observer. Error bounds and robustness to uncertainties in the models and in the inputs are found. The estimation is independent of the kinetic models of the microorganism. The characteristic features of the observer are illustrated by numerical simulations and validated by experimental results. [ABSTRACT FROM AUTHOR]

Water industry is facing increased pressure to produce higher quality treated water at a lower cost. The coagulation-flocculation unit is a major step in the production of potable water, allowing the removal of colloidal particles and contamination sources. Currently, the coagulant dosing is manual and it is determined by the way of jar test (off-line dosing tests). This test cannot be applied to supervision and diagnostic real-time. In order to obtain a simple model to describe the water treatment plant, a behavior model sets out, from the analysis of raw water characteristics to the entrance of the plant: (1) prediction on-line the amount of optimal coagulant dosage (software sensor) and (2) the determination of the functional states in real time. The innovative aspect of this paper resides in the design of intelligent control techniques of information water treatment plant (fuzzy logic for the classifier and neuronal networks for the software sensor). [ABSTRACT FROM PUBLISHER]

The lack of sensors for some relevant state variables in fermentation processes can be coped by developing appropriate software sensors. In this work, NARX-ANN, NARMAX-ANN, NARX-SVM and NARMAX-SVM models are compared when acting as software sensors of biomass concentration for a solid substrate cultivation (SSC) process. Results show that NARMAX-SVM outperforms the other models with an SMAPE index under 9 for a 20 % amplitude noise. In addition, NARMAX models perform better than NARX models under the same noise conditions because of their better predictive capabilities as they include prediction errors as inputs. In the case of perturbation of initial conditions of the autoregressive variable, NARX models exhibited better convergence capabilities. This work also confirms that a difficult to measure variable, like biomass concentration, can be estimated on-line from easy to measure variables like CO and O using an adequate software sensor based on computational intelligence techniques. [ABSTRACT FROM AUTHOR]

Abstract: Specific kinetic rates are key variables regarding metabolic activity in bioprocesses. They are non-linear functions of concentrations and operating conditions and therefore of difficult access for process control. In this paper, a multiple kinetic rates observer based on second-order sliding mode ideas is proposed. The main difference with other proposals is that smooth estimates are achieved in finite-time without adding additional dynamics. The resulting estimator is robust against uncertainty in the model of the estimated variables. Experimental results from continuous fermentation of S. cerevisiae are presented, where microbial specific growth rate and net ethanol production rate are estimated. [Copyright &y& Elsevier]

Abstract: Data-driven modeling techniques are applied to process data from wastewater treatment plants to provide valuable additional information for optimal plant control. The application of data-driven modeling techniques, however, bears some risk because the generated models are of non-mechanistic nature and they thus do not always describe the plant processes appropriately. In this study, a procedure to build software sensors based on sensor data available in the process information system is defined and used to compare several techniques suitable for data-driven modeling, including generalized least squares regression, artificial neural networks, self-organizing maps and random forests. Three different degrees of expert knowledge are defined and considered mainly for optimum input signal selection and model interpretation. In two full-scale experiments, software sensors are created. The experiments reveal that even with linear modeling techniques, it is possible to automatically generate accurate software sensors. Hence, this justifies the selection of the most parsimonious and transparent models and to motivate their investigation by taking into account available expert knowledge. A high degree of expert knowledge is valuable for long-term accuracy, but can lead to performance decreases in short-term predictions. With regard to safe on-site deployment, the consideration of uncertainty measures is crucial to prevent misinterpretation of software-sensor outputs in the cases of rare events or model input failures. [Copyright &y& Elsevier]

Abstract: Biological and environmental systems are often influenced by unknown inputs or disturbances, which makes monitoring or state estimation more delicate. In this study, the simultaneous estimation of unmeasured state variables and partly unknown inputs is considered. Only qualitative prior information on these inputs is used in the design procedure, leading to the concept of quasi-unknown input observers (QUIO). These software sensors are applied to the estimation of concentrations, flow rates and light intensity in phytoplanktonic cultures in the chemostat. Implementation and numerical tests are discussed, based on simulation and experimental data. [Copyright &y& Elsevier]

Abstract: The present study is dedicated to the estimation of internal temperatures within a foodstuff during microwave tempering, in the presence of unknown dielectric properties. The solution consists of a software sensor based on a model originating from the closed-form solutions of Maxwell''s equations coupled with the conduction heat equation solved by finite differences. The algorithm implemented enables convergence towards acceptable dielectric property functions while the temperature field is estimated simultaneously. This approach is carried out in simulation by considering the tempering of a block of raw beef located in a rectangular wave guide in order to consider a fundamental mode with perfectly known electromagnetic conditions. Industrial Relevance: In the industrial processes, the lack of sensors constitutes a brake to the design of relevant control strategies. One well-known solution is to implement software sensors (also called observers) to estimate online the lacking data. However, such a solution requires a good knowledge of the dynamical model structure and the model parameters. The development of software sensors remains nevertheless a difficult task for non specialist in process control. CLPP presented in this submission is particularly well suited to industrial cases, because it permits to estimate with a good accuracy the lacking state variables in the presence of large uncertainties on the parameters'' vector. The interest is illustrated on a microwave food tempering process, where dielectric parameters are unknown, and where the available measurements are only surface temperatures. [Copyright &y& Elsevier]

Abstract: In this study two approaches to predict the total alkalinity (expressed as mgL−1 ) of an anaerobic digester are examined: firstly, software sensors based on multiple linear regression algorithms using data from pH, redox potential and electrical conductivity and secondly, near infrared reflectance spectroscopy (NIRS). Of the software sensors, the model using data from all three probes but a smaller dataset using total alkalinity values below 6000mgL−1 produced the best calibration model (R 2 =0.76 and root mean square error of prediction (RMSEP) of 969mgL−1 ). When validated with new data, the NIRS method produced the best model (R 2 =0.87 RMSEP=1230mgL−1 ). The NIRS sensor correlated better with new data (R 2 =0.54). In conclusion, this study has developed new and improved algorithms for monitoring total alkalinity within anaerobic digestion systems which will facilitate real-time optimisation of methane production. [Copyright &y& Elsevier]

This study proposes two adaptive control algorithms for the fed-batch production of α-amylase. The first one uses online information from hardware measuring glucose. Online information of both biomass and glucose concentrations measured with different frequency is used in the second algorithm. Hardware measuring variables are inputs for software sensors of glucose concentration and (specific) glucose consumption rate. Either of the algorithms do not require any kinetic coefficients. This is a benefit, because the kinetic coefficients can vary during cultivation and between cultivations, leading to low process reproducibility and the non-stationary state of the bioprocess. The results of simulation investigations show good performance of the proposed control schemes. [ABSTRACT FROM AUTHOR]