Local Linear Model Tree Research Articles

System modeling and temperature control for a fuel cell system based on local model networks

ABSTRACT Fuel cells have been studied for use in stationary power generation and vehicle propulsion systems. The fuel cell thermal management subsystem is coupled and nonlinear, posing challenges for modeling and temperature control. This paper aims to integrate the physical models of the fuel cell stack, pump, thermostat, and other components combined with intelligent algorithms into an efficient system-level thermal management model framework and develop a model predictive controller to solve the temperature control problem. First, a physics-based nonlinear model of the fuel cell system is developed and used as a basis to identify the linearized model for different operating points. Then, the global model is obtained by fusing the local models with Gaussian validity functions using the local linear model tree method. Third, a multi-step prediction model is derived based on the local model networks, and a parameterized linear state space form is obtained and used for controller design. Furthermore, an online correction method is developed to reduce the model discrepancy. Finally, the accuracy of the system model and the performance of the proposed controller are verified by open-loop experimental data and a series of closed-loop simulation cases.

Structurally Optimized Neural Fuzzy Modeling for Model Predictive Control

This paper investigates the local linear model tree (LOLIMOT), a typical neural fuzzy model, in the multiple-input-multiple-output model predictive control (MPC). In the conventional LOLIMOT, the structural parameters including centres and variances of its Gaussian kernels are set based on equally dividing the input data space. In this paper, after the structural parameters are initially obtained from the input space partition, they are optimized by the gradient descent search, from which the space partitions are further adjusted. This makes it better for the model structure to fit the input data statistics, leading to improved modelling performance with small model size. The MPC based on the proposed structurally optimized LOLIMOT is then implemented and verified with both numerical and diesel engine plants. Validation results show that the proposed MPC has significantly better controlling performance than the MPC based on the conventional LOLIMOT, making it an attractive solution in practice.

Control oriented modeling of twin-screw granulation in the ConsiGmaTM-25 production plant

ConsiGmaTM-25 is a continuous production plant integrating a twin-screw granulation, fluid bed drying, granule conditioning, and a tableting unit. The particle size distribution (PSD), active pharmaceutical ingredient (API) content, and liquid content of wet granules after twin-screw granulation affect the quality of intermediate and final products. This paper proposes methods for real-time monitoring of these quantities and control-oriented modeling of the granulator.The PSD of wet granules is monitored via an in-line process analytical technology (PAT) probe based on the spatial velocimetry principle. The algorithm for signal processing and evaluation of PSD characteristics is developed and applied to the acquired PSD data. A dynamic process model predicting PSD characteristics from granulation parameters is trained via the local linear model tree (LoLiMoT) approach. The experimental data required for the model training are collected via systematically designed excitation runs. Finally, the performance of the identified model is examined and verified by means of a new set of validation runs. Furthermore, an in-line PAT probe based on Raman spectroscopy is developed and integrated after the granulator. The API- and liquid content of produced wet granules are evaluated from the spectral data by means of chemometric modeling, and chemometric models are validated on a separate set of experimental data. The solutions proposed in this research can be used as a reliable (and necessary) basis for the development of advanced quality-by-design control concepts (e.g., PSD process control). Such concepts would ultimately improve the ConsiGmaTM-25 process performance in terms of robustness against disturbances and quality of intermediate and final products.

Retracted: Estimating Lane Change Duration for Overtaking in Nonlane‐Based Driving Behavior by Local Linear Model Trees (LOLIMOT)

Retracted: Estimating Lane Change Duration for Overtaking in Nonlane‐Based Driving Behavior by Local Linear Model Trees (LOLIMOT)

Estimating Lane Change Duration for Overtaking in Nonlane-Based Driving Behavior by Local Linear Model Trees (LOLIMOT)

Lane change (LC) is one of the main maneuvers in traffic flow. Many studies have estimated LC duration directly by using lane-based data. The current research presents an estimate of LC duration for overtaking maneuver in nonlane-based traffic flow. In this paper, the LC duration is estimated implicitly by modeling lateral speed and applying the length of required lateral movement to complete the LC maneuver. In lateral speed modeling, the local linear model tree is applied which consists of three variables: the initial lateral distance, longitudinal speed, and time to collision (TTC), which itself is a function of the relative speed of follower and the distance between the two vehicles. The initial lateral distance is the relative transverse distance from which the following vehicle initializes the LC. The range of lateral speed was estimated between 0.5 and 5 km/h, which resulted in the LC duration between 2.5 and 24 sec. The results indicate that the lateral and longitudinal speed would be inversely related, while the lateral speed and the initial transverse distance as well as TTC would be directly related. The findings also indicate that TTC can be assumed as the most important factor affecting lateral speed. TTC at 8 sec can be considered as the threshold for its effect on the LC duration since at longer TTCs, and the lateral speed has remained almost constant. When TTC is longer than 8 sec, it would not affect the LC duration.

Optimized hybrid decoupled visual servoing with supervised learning

This study proposes an optimized hybrid visual servoing approach to overcome the imperfections of classical two-dimensional, three-dimensional and hybrid visual servoing methods. These imperfections are mostly convergence issues, non-optimized trajectories, expensive calculations and singularities. The proposed method provides more efficient optimized trajectories with shorter camera path for the robot than image-based and classical hybrid visual servoing methods. Moreover, it is less likely to lose the object from the camera field of view, and it is more robust to camera calibration than the classical position-based and hybrid visual servoing methods. The drawbacks in two-dimensional visual servoing are mostly related to the camera retreat and rotational motions. To tackle these drawbacks, rotations and translations in Z-axis have been separately controlled from three-dimensional estimation of the visual features. The pseudo-inverse of the proposed interaction matrix is approximated by a neuro-fuzzy neural network called local linear model tree. Using local linear model tree, the controller avoids the singularities and ill-conditioning of the proposed interaction matrix and makes it robust to image noises and camera parameters. The proposed method has been compared with classical image-based, position-based and hybrid visual servoing methods, both in simulation and in the real world using a 7-degree-of-freedom arm robot.

Robust outlier detection in geo-spatial data based on LOLIMOT and KNN search

One of the most challenging topics in analyzing multi-dimensional geo-spatial data such as geophysical data-sets is detecting outlier data. The issue mainly originates from the difficulty in describing “normality” or “abnormality” due to the complexity of the relationships between the data elements. Considerable number of methods have been proposed and applied for detecting outliers whether they are assumed to be noise, anomalies within the data-set or simply isolated events. A new outlier detection method reached from automatic training of Local Linear Model Tree (LOLIMOT) network, and based on the data selected by K-Nearest Neighborhood (KNN) search is proposed in this research. The procedure of selecting data pairs is through decile analysis using distances calculated during KNN data grouping. Experiment on a synthetic 12 cluster 3D data-set is indicative of the method’s robust performance where calculated Cumulative Error Percentage (CEP) is 13% for the method whereas the nearest value for the KNN is 19%. Also, by applying the method on a micro-gravimetric data and an earthquake catalogue related to the north Zagros- west Alborz, and based on the output of the analyses performed, the superiority of the method in outlier detection was confirmed.

Railway crossing vertical vibration response prediction using a data-driven neuro-fuzzy model – Influence of train factors

This paper provides a data-driven model of the vibration response of a railway crossing during vehicle passages. Many of the features of trains passing through instrumented crossing are extracted from measured data. Based on the feature selection process, speed, dynamic axle load and the number of wagons are found proper inputs in the prediction model. Train-crossing interaction response at a crossing due to passing trains is modeled from a data-driven Neuro-Fuzzy soft computing approach. Locally Linear Model Tree (LOLIMOT) is applied to predict the crossing nose acceleration. The model comparison against measurements shows that the ability to predict the extrapolation cases at off-range speeds has satisfactory compatibility. The monitored passing trains are ranked based on the LOLIMOT input space dimension cuts and extrapolation of the model up to higher train speeds. The influence of train factors (i.e. speed, dynamic axle load, number of wagons) on crossing response is demonstrated. Also, based on the analysis results, it is concluded that with a steady increase in train speeds, some trains show a greater amplification in vibration response than others. The results can be applied in data processing in the crossing vibration monitoring and detection of trains with crossing impact sensitive to speed increasing that can lead to proper operation policies to reduce damages and maintenance costs.

Extracting Inherent Model Structures and Identifying Parameters of Time-Varying Systems Using Local Linear Neuro-Fuzzy Networks

The main challenges in identifying time-varying systems are to extract the inherent model structure and formulate a generic approach for various time-varying processes. In this article, a local linear neuro-fuzzy networks based approach is proposed for the first time to tackle this issue. The basic idea is to utilize the superior approximation capability of the fuzzy neural networks to expand the unknown time-varying parameters so that various nonlinear and discontinuous processes can be captured. In addition, the local linear model tree optimization algorithm is incorporated in our approach to optimize the network architecture. On one hand, the least number of neurons corresponding to the time-varying parameters can be automatically determined, thus avoiding the superfluous bases and the troublesome overfitting problem; on the other hand, the optimized distribution of the neurons could significantly increase the interpretability of the results, thus contributing to locating the domains where the parameters exhibit strong nonlinearity or change sharply. Moreover, we demonstrate that our approach could simultaneously extract the parsimonious model structure while identifying the time-varying parameters, without adding additional procedures or increasing computation burden. Considering the generality and the unique merits, our proposed approach could significantly advance the applications of artificial neural network techniques in system identification.

Estimation of porosity of rocks based upon axiomatic local linear model tree (a lolimot model) approach

Shear and Compressional Wave Velocities along with other Petrophysical Logs, are considered as upmost important data for Hydrocarbon reservoirs characterization. In this study, porosity of the extracted rocks form concerned wells is interest as it can indicate the oil capacity of the wells of interest. In this study, we employ the principles of Axiomatic Design theory, specially the first (independence) axiom, to more simplify the measurement system. Also, to clarify the strength of Axiomatic Design theory in reducing the complexity of the system and optimizing the measurement system, we utilize the The Lolimot model (LOcal LInear MOdel Tree) as a model from the neural network family and apply it before and after implementing the basic logic of Axiomatic Design (AD) theory. In addition, in order to illustrate strength of the proposed method emphasizing the effectiveness of a method which benefit from both AD theory and Lolimot model together, the existing system used to measure the rock porosity is addressed and actual data related to one of wells located in southern Iran is utilized. The results of the study show that integrating the Axiomatic Design principles with the LOLIMOT method leads to the least complex and most accurate results.

Classification-Based Fuel Injection Fault Detection of a Trainset Diesel Engine Using Vibration Signature Analysis

Abstract Diesel engines are crucial components of trainsets. Automated fault detection of diesel engines can play an important role for increasing reliability of passenger trains. In this research, vibration-based fuel injection fault detection of a high-power 12-cylinder trainset diesel engine is studied. Vibration signals are analyzed in frequency and time-frequency domains to obtain possible patterns of faults. Fast Fourier transform (FFT) and wavelet packet transform (WPT) of vibration signals are used to extract several uncorrelated features. These features are chosen to increase the ability of classifiers to separate healthy and faulty engine sides, automatically. Different classification methods including multilayer perception (MLP), support vector machines (SVM), K-nearest neighbor (KNN), and local linear model tree (LOLIMOT) are used to process captured features; these methods are utilized in both “Single-sensor condition monitoring” and “Classification and fault detection” sections. It is shown that KNN networks are practical tools in the proposed fault detection procedure. The main novelty of this work comes from introducing a rich feature-extraction method based on a combination of FFT and db4 features. In addition, the complexity of computations and average running-time decrease while classification accuracy in the fuel injection fault detection procedure increases.

Short-term load forecast using ensemble neuro-fuzzy model

In this paper, Takagi-Sugeno-Kang neuro-fuzzy model is trained using locally linear model tree (LOLIMOT) method to forecast day-ahead hourly load profile. The proposed approach is applied to a real load profile measured in Iran as a geographically spread case study. The effects of partitioning the power system to smaller regions on the load forecasting and its advantages, such as practical consideration of daily average temperature data, are also shown. Moreover, a set of preprocessing approaches is proposed and implemented on historical load data to improve forecasting results. It is shown that by using LOLIMOT, the neuro-fuzzy model does not need the predetermined settings, such as the number of neurons, membership functions or fuzzy rules by an expert because all the parameters are set by the LOLIMOT method. This approach leads to the flexible network topology of the trained model for different days, which leads to extract the load profile trends more effectively.

Identification of Geochemical Anomalies Using Fractal and LOLIMOT Neuro-Fuzzy modeling in Mial Area, Central Iran

The Urumieh-Dokhtar Magmatic Arc (UDMA) is recognized as an important porphyry, disseminated, vein-type and polymetallic mineralization arc. The aim of this study is to identify and subsequently determine geochemical anomalies for exploration of Pb, Zn and Cu mineralization in Mial district situated in UDMA. Factor analysis, Concentration-Number (C-N) fractal model and Local Linear Model Tree (LOLIMOT) algorithm used for this purpose. Factor analysis utilized in recognition of the correlation between elements and their classification. This classified data used for training the LOLIMOT algorithm based on relevant elements. The results of the LOLIMOT algorithm represent anomalies in areas with no lithogeochemical samples. Although, the C-N log-log plot for target elements were generated based on stream sediment and lithogeochemical samples which could be delineated mineral potential maps of the target elements. Results obtained by the LOLIMOT and fractal modeling show that the SW and the Eastern parts of the area are proper for further exploration of Cu, Pb, and Zn.

Machine learning methods for optimal prediction of motor outcome in Parkinson’s disease

PurposeIt is vital to appropriately power clinical trials towards discovery of novel disease-modifying therapies for Parkinson’s disease (PD). Thus, it is critical to improve prediction of outcome in PD patients. MethodsWe systematically probed a range of robust predictor algorithms, aiming to find best combinations of features for significantly improved prediction of motor outcome (MDS-UPDRS-III) in PD. We analyzed 204 PD patients with 18 features (clinical measures; dopamine-transporter (DAT) SPECT imaging measures), performing different randomized arrangements and utilizing data from 64%/6%/30% of patients in each arrangement for training/training validation/final testing. We pursued 3 approaches: i) 10 predictor algorithms (accompanied with automated machine learning hyperparameter tuning) were first applied on 32 experimentally created combinations of 18 features, ii) we utilized Feature Subset Selector Algorithms (FSSAs) for more systematic initial feature selection, and iii) considered all possible combinations between 18 features (262,143 states) to assess contributions of individual features. ResultsA specific set (set 18) applied to the LOLIMOT (Local Linear Model Trees) predictor machine resulted in the lowest absolute error 4.32 ± 0.19, when we firstly experimentally created 32 combinations of 18 features. Subsequently, 2 FSSAs (Genetic Algorithm (GA) and Ant Colony Optimization (ACO)) selecting 5 features, combined with LOLIMOT, reached an error of 4.15 ± 0.46. Our final analysis indicated that longitudinal motor measures (MDS-UPDRS-III years 0 and 1) were highly significant predictors of motor outcome. ConclusionsWe demonstrate excellent prediction of motor outcome in PD patients by employing automated hyperparameter tuning and optimal utilization of FSSAs and predictor algorithms.

Local Linear Model Tree with Optimized Structure

This paper investigates the local linear model tree (LOLIMOT) with optimized structure. The performance of the LOLIMOT model depends on how the neurons are constructed. In the typical LOLIMOT model, the number of neurons is initially set as one and starts to increase by repeatedly splitting an existing neuron into two equal ones until the required performance is achieved. Because the equal split of a neuron is not optimal, a large model size is often necessary for required performance, leading to high complexity and strong overfitting. In this paper, we propose a gradient-decent-search-based algorithm to optimally split an existing neuron into two new ones. Based on both numerical data and simulated engine data, through the evaluation of optimized structure, the effectiveness of proposed method has been verified.

Data-Driven Modelling of the Nonlinear Cortical Responses Generated by Continuous Mechanical Perturbations

Cortical responses to external mechanical stimuli recorded by electroencephalography have demonstrated complex nonlinearity with fast dynamics. Hence, the modelling of the human nervous system plays a crucial role in studying the function of the sensorimotor system and can help in disentangling the sensory-motor abnormalities in functional movement disorders. In this paper, a non-parametric model is estimated based on locally-linear neuro-fuzzy structures trained by an evolutive algorithm relying on locally-linear model-tree. In particular, simulation model as well as a multi-step ahead predictor model is considered to describe the nonlinear dynamics governing the cortical response. The proposed modelling method is applied to an experimental dataset representing brain activities from ten young healthy subjects. These electroencephalography signals are recorded while robotic manipulations have been applied to their wrist joints. The obtained results are satisfactory and are also compared to those achieved with different modelling strategies applied to the same benchmark data.

Improving Pellet Quality in a Pharmaceutical Hot Melt Extrusion Process via PID Control and LOLIMOT-Based MPC

PurposeThe aim of this paper is the development of a process control concept for a hot melt extrusion (HME) and pelletization process. The new concept should improve the particle size distribution of the pellets produced.MethodsProduction of pellets containing an active pharmaceutical ingredient (API) can be achieved by means of HME, followed by a pelletization process step. The quality of pellets produced depends on the strand temperature at the pelletizer’s inlet and the pelletizer’s intake speed. This paper presents a strategy for the strand diameter and temperature control based on adjusting the cooling intensity on the cooling track between the HME and the pelletization step and altering the pelletizer’s intake speed. Two concepts are presented and compared to the open-loop operation of the system: the first one is model predictive control (MPC) in combination with a model based on the local linear model tree (LOLIMOT) algorithm, and the second one is PID control. The quality of the pellets produced was analyzed in terms of particle size distribution (PSD).ResultsBy implementation of the two control concepts, strand temperature and diameter could be kept close to the desired set points. Consequently, the presented concepts yielded pellets with a narrower particle size distribution than the open-loop operation of the plant.ConclusionsThe application of the presented control strategies can improve the quality of the pellets produced by an HME and pelletization system in terms of their particle size distribution.

Neuro-Fuzzy Modeling of Data Singular Spectrum Decomposition and Traffic Flow Prediction

Dynamic traffic flow prediction is one of the important topics in traffic engineering and intelligent transportation systems. The traffic flow has stochastic nature and nonlinear dynamics which make accurate prediction of traffic flow a challenging process. In most existing methods, the nonlinear and stochastic qualities of traffic flow have not been taken into consideration simultaneously and these methods do not possess satisfactory accuracy either. In this paper, in order to reduce the prediction error based on traffic flow characteristics, we applied two different methods. At first, we applied a locally linear neuro-fuzzy method using local linear model tree learning algorithm for nonlinear identification of traffic flow. Then, to eliminate the noisy components and also to increase the prediction accuracy, we proposed a method, which is the combination of singular spectrum analysis (SSA) and locally linear neuro-fuzzy model (LLNF). In this method, firstly, the principal components of the traffic flow time series were extracted, several noisy and unimportant components were thrown out and then every remained important component was modeled by using a LLNF network, the trained networks were utilized for one-step-ahead prediction, and finally the predicted patterns were combined to construct the general prediction. Moreover, we compared the results of these two methods with each other and also with those of several other intelligent methods such as multi-layer perceptron, radial-basis function, network and adaptive network fuzzy inference system. The simulation results revealed that the proposed SSA + LLNF approach for traffic flow prediction had promising and superior performance.

Optimized machine learning methods for prediction of cognitive outcome in Parkinson's disease

BackgroundGiven the increasing recognition of the significance of non-motor symptoms in Parkinson's disease, we investigate the optimal use of machine learning methods for the prediction of the Montreal Cognitive Assessment (MoCA) score at year 4 from longitudinal data obtained at years 0 and 1. MethodsWe selected n = 184 PD subjects from the Parkinson's Progressive Marker Initiative (PPMI) database (93 features). A range of robust predictor algorithms (accompanied with automated machine learning hyperparameter tuning) and feature subset selector algorithms (FSSAs) were selected. We utilized 65%, 5% and 30% of patients in each arrangement for training, training validation and final testing respectively (10 randomized arrangements). For further testing, we enrolled 308 additional patients. ResultsFirst, we employed 10 predictor algorithms, provided with all 93 features; an error of 1.83 ± 0.13 was obtained by LASSOLAR (Least Absolute Shrinkage and Selection Operator - Least Angle Regression). Subsequently, we used feature subset selection followed by predictor algorithms. GA (Genetic Algorithm) selected 18 features; subsequently LOLIMOT (Local Linear Model Trees) reached an error of 1.70 ± 0.10. DE (Differential evolution) also selected 18 features and coupled with Thiel-Sen regression arrived at a similar performance. NSGAII (Non-dominated sorting genetic algorithm) yielded the best performance: it selected six vital features, which combined with LOLIMOT reached an error of 1.68 ± 0.12. Finally, using this last approach on independent test data, we reached an error of 1.65. ConclusionBy employing appropriate optimization tools (including automated hyperparameter tuning), it is possible to improve prediction of cognitive outcome. Overall, we conclude that optimal utilization of FSSAs and predictor algorithms can produce very good prediction of cognitive outcome in PD patients.

Identification of piecewise linear dynamical systems using physically-interpretable neural-fuzzy networks: Methods and applications to origami structures

Self-locking origami structures are characterized by their piecewise linear constitutive relations between force and deformation, which, in practice, are always completely opaque and unmeasurable: the number of piecewise segments, the positions of non-smooth points, and the linear parameters of each segment are unknown a priori. However, acquiring this information is of fundamental importance for understanding the origami structure’s dynamic folding process and predicting its dynamic behaviors. This, therefore, arouses our interest in adopting a dynamical identification process to determine the model and to estimate the parameters. In this research, based on the piecewise linear assumption, a physically-interpretable neural-fuzzy network is built to correlate the measured input and output data. Unlike the conventional approaches, the constructed neural network possesses specific physical meaning of its components: the number of neurons relates to the number of piecewise segments, the coefficients of the local linear models relate to the parameters of the constitutive relations, and the validity functions relate to the positions of non-smooth points. By addressing several examples with different backgrounds, the network’s underlying data training methods are illustrated, including the local linear optimization for linear parameters, nested optimization for nonlinear partitions, and Local Linear Model Tree optimization for model selection. Noting that the tackled origami problem holds strong universality in terms of the unknown piecewise characteristics, the proposed approach would thus provide an effective, generic, and physically significant means for handling piecewise linear dynamical systems and meanwhile bring fresh vitality to the artificial neural network research.