International Journal of Computational Methods

This work develops an automatic lawnmower (Auto-Lawnmower) using computer vision technology for obstacle avoidance. Several critical issues have been overcome in this development work, including the development of a simplified convolutional neural network (CNN) for decision making, and sufficiently large datasets needed to train the Auto-Lawnmower. The following key strategies are adapted to ensure necessary functionality and efficacy with minimum cost, considering possible mass production of Auto-Lawnmowers. First, we use at-time avoidance strategy: meaning the Auto-Lawnmower makes turns when it encounters an obstacle. This is possible because of the fact that a lawn mower usually can move at a very low speed (walking speed of a man). Second, we decided to use minimum necessary sensors, so as to make the system of our Auto-Lawnmower as simple as possible and hence can be very practical. A single monocular video camera is, therefore, used as the single sensor for both: (1) to generate real-time movie images of a large number of situations that the Auto-Lawnmower may encounter, which is labelled and builds up the training datasets; (2) to capture the real-time image when the trained Auto-Lawnmower is in action, to decide on either to avoid the obstacle or move forward while cutting the grass. A Law dataset with labels, called LawNet, has been established for training of CNNs. It collects a total of 168,542 labelled images taken from the perspective of our Auto-Lawnmower for lawns at the university campus. A concise CNN is created for high efficiency and trained with our LawNet to drive the Auto-Lawnmower. A prototype of Auto-Lawnmower is finally designed, built and tested for lawn mowing in real situations. It is found that it works well as designed, with minimum sensors (a single camera), and hence it has a good potential for mass production.

Symbolic regression (SR), as a special machine learning method, can produce mathematical models with explicit expressions. It has received increasing attention in recent years. However, finding a concise, accurate expression is still challenging because of its huge search space. In this work, a divide and conquer (D&C) scheme is proposed. It tries to divide the search space into a number of orthogonal sub-spaces based on the separability feature inferred from the sample data (dividing process). For each sub-space, a sub-function is learned (conquering process). The target model function is then reconstructed with the sub-functions according to their separability patterns. To this end, a separability pattern detecting technique, bi-correlation test (Bi-CT), is also proposed. Note that the sub-functions could be determined by any of the existing SR methods, which makes D&C easy to use. The D&C powered SR has been tested on many symbolic regression problems, and the study shows that D&C can help SR to get the target function more quickly and reliably.

Hydraulic fracture propagation directly affects the recovery rate of resources when hydraulic fracturing techniques are applied to exploiting unconventional oil and gas resources. Rock mass is the main engineering medium of hydraulic fractures and is generally considered to be of considerable spatial variability in physical and mechanical properties. Understanding the irregular propagation mechanism of hydraulic fracture in spatial heterogeneity rock mass is essential and beneficial to assess the recovery rate of oil or gas resources. This work develops a random phase-field method (RPFM) to simulate the irregular propagation of hydraulic fracture in spatially variable rock mass. The spatial variability of elastic modulus is characterized by the random field theory. Utilizing the advantages in modeling complex crack patterns and crack kinematics, the phase-field method (PFM) is used to predict the fracture propagation. Various anisotropic random fields of elastic modulus with different coefficients of variance and scales of fluctuation are generated via the Cholesky decomposition method. The random fields are subsequently implemented into COMSOL Multiphysics and combined with the PFM to investigate the hydraulic fracture propagation. This study investigates the influence of spatial variability of elastic modulus on the peak fluid pressure, fracture length, fracture area and fracture shape. It reveals that the spatial variability of elastic modulus has a significant influence on the propagation of hydraulic fractures, and the results provide a preliminary reference for hydraulic fracturing design with consideration of spatial variability of rock mass.

The study and prediction of soil liquefaction is an important and complex issue in geotechnical earthquake engineering. This paper attempts to compare the predictability of soil liquefaction potential between several machine learning classification models, which includes some tree-based classifiers, multilayer perceptron (MLP) neural networks, Support Vector Machine (SVM), some state-of-the-art ensemble methods, K nearest neighbors method, classical Naive Bayesian classifier and logistic regression. Three data sets covering shear-wave velocity measurements, cone penetration testing (CPT), and real historic earthquakes cases are employed to train and evaluate the machine learning classifiers. In order to make the best use of large varieties of statistical and machine learning classification algorithms, it is necessary to give a comparative evaluation of the model performance before model selection and offer advice on a unified stable model for all sorts of collected datasets. In the comparative study, data preprocessing is first performed to ensure the dataset into all machine models is of good quality. Then all three datasets with different input features are passed into the machine learning algorithms to obtain its confusion matrix and some evaluation indices. Reliable assessment of model performance is done through a repeated sub-sampling process. Experimental results are also supported by ROC curves. The results of this study indicated that although most machine learning methods are able to represent the complex relationship between seismic proper seismic properties of soils and corresponding liquefaction potential, ensemble learning has achieved more successful results in all three datasets test and can be a fairly promising approach on prediction of earthquake-induced soil liquefaction.

Speech signal often gets corrupted by different noises like airport noise, station noise, and street noise. These noises tend to degrade the quality of the speech signal, particularly in voice communication, automatic speech recognition, and speaker identification. Therefore, it is necessary for automatic speech enhancement. In this research work, a novel speech signal enhancement model is introduced with the assistance of deep learning. The proposed model includes three major phases: (a) pre-processing, (b) feature extraction, and (c) speech enhancement. In the pre-processing phase, the framing will be carried out using the Hanning window, where the input speech signals will be decomposed into a series of overlapping frames. Then, from these individual frames, the multi-features like the improved Mel-frequency cepstral coefficients (IMFCCs), fractional delta AMS, and modified STFT (M-STFT) will be extracted. Subsequently, in the speech enhancement phase, the available noise is estimated initially, and it is removed. The noise removed signals from the frames are used to determine the optimal mask of all the frames of the noisy speech signal, and the mask is employed for training the Deep Convolutional Neural Network (DCNN). The reconstructed outcomes from DCNN are the enhanced speech signal. Finally, the proposed work (multi-features+ DCNN-based Speech Enhancement) is validated over existing models in terms of certain measures, which exhibits the supremacy of the proposed work.

Wireless Sensor Network (WSN) localization has been bloomed as an active area of research in this era. In fact, WSN differs from the traditional network in diverse aspects and therefore requires novel algorithms for addressing specific challenges like the identification of the unknown node location in hazardous environments. In this paper, a new localization model is introduced by the range-based localization approach via an optimization-assisted deep learning model. The proposed work undergoes two major phases: (a) training phase and (b) localization phase. The trained Deep Neural Network (DNN) with the measured distance-based features like the “Angle of Arrival (AoA) and Received Signal Strength Indicator (RSSI)” find out the place of the unknown node more precise. Further, to enhance the localization accuracy, the weight of DNN is tuned via a novel hybrid optimization algorithm named as Lion-Assisted Firefly Algorithm (LAFA) model. The proposed LAFA is the concept of both the Lion Algorithm (LA) and Firefly Algorithm (FF). Finally, the evaluation of the presented work is done in terms of error measures.

The brain tumor is the most serious cancer among people of all ages, and recognition of its grade is a complex task for monitoring health. In addition, the earlier detection and classification of tumors into a particular grade are imperative for diagnosing the tumor effectively. This paper devises a novel method for multigrade tumor classification using deep architecture. First, the pre-processing is performed with the Region of interest (ROI) and Type 2 Fuzzy and Cuckoo Search (T2FCS) filter. After that, segmentation using a pre-processed image is carried out to generate segments, which is performed using a deep fuzzy clustering model. Then, the significant features are mined through segments that involve convolution neural network (CNN) features, Texton features, EMD features, and statistical features such as mean, variance, kurtosis, and entropy. The obtained features are subjected to Deep Residual Network for multigrade tumor classification. The Deep Residual Network training is done with the proposed Harmony search-based Feedback Artificial Tree (HSFAT) algorithm. The proposed HSFAT is devised by combining Harmony search and Feedback Artificial Tree (FAT) algorithm. The proposed HSFAT-based deep residual network provided superior performance with maximum accuracy of 94.33%, maximum sensitivity of 97.27%, and maximum specificity of 92.61%.

Over the last several decades, the Finite Element Method (FEM) has emerged as a numerical approach method of choice for the solution of problems in solid mechanics. Part of the reason for the success of FEM is that it provides a unified framework for discretizing even complex differential equations. However, despite this overall unification, FEM still requires specific variants or corrections depending on the problem at hand. For instance, problems with skewed meshes, discontinuity, singularity, incompressible media, etc. require the analyst to modify the discretization approach in order to preserve robustness. We speculate that local-polynomial bases such as those used in FEM do not sufficiently represent local physics and more “physics-informed” approaches may be more universal. Accordingly, in this paper, we evaluate the feasibility of one such approach — the recently developed Distributed Physics Informed Neural Network (DPINN) approach — to provide a truly unified framework for addressing problems in Solid Mechanics. The DPINN approach utilizes a piecewise-neural network representation for the underlying field, rather than the piece-polynomial representation that is common in FEM. We solve a series of problems in solid mechanics using either the single or domain-distributed version of DPINN and demonstrate that the approach is able to seamlessly solve varied problems with no special treatment required for volumetric locking or capturing discontinuities. Further, we also demonstrate that the DPINN approach, due to its meshless nature, is able to avoid the curse of dimensionality. We discuss the relative merits and demerits of the DPINN approach in comparison to FEM. We expect this work to be useful to researchers looking to develop unified computational frameworks for problems in solid mechanics.

Stewart parallel manipulator is well known for its superiority in achieving better stiffness, accurate motion with precise positioning, robust mechanism, high payload capacity, etc. It is widely used in various applications such as flight simulators, satellite dish positioning, hexapod telescope, medical surgery, simulation of earthquakes, etc. It is important to note that the inverse kinematics solution of the Stewart platform can be determined easily with the help of an analytical solution, whereas forward kinematics is intractable analytically. Therefore, in this work, an attempt is made to solve the forward kinematics problem of the Stewart platform using the soft-computing-based technique. A multi-layer feed-forward neural network with one hidden layer is trained after utilizing different metaheuristic optimizers, namely Particle Swarm Optimization (PSO), Modified Chaotic, Invasive Weed Optimization (MCIWO), and Teachers’ Learning-Based Optimization (TLBO) methodologies to solve the forward kinematics of the Stewart platform. Further, a detailed analysis is conducted on the results obtained by these methods, namely PSO-NN, MCIWO-NN and TLBO-NN. The dataset for training the NN is generated by using the solution of inverse kinematics.

This research article attempts to design Inverse Dynamics Controller (IDC) to execute the motion tracking of Stewart Platform. In the presence of modeling uncertainties and external disturbances, the closed-loop dynamic equation of IDC with fixed gains becomes nonlinear and configuration-dependent, which compromises the motion tracking accuracy. Further, both modeling uncertainties and external disturbances are unavoidable in real life conditions. To tackle this issue, this article proposes a novel control algorithm by combining IDC with Feed-Forward Artificial Neural Network (FF-ANN) trained using PSO. The proposed modified control algorithm offers superior motion tracking accuracy in comparison with traditional IDC.

Fire-loaded concrete structures may experience explosive spalling, i.e., violent splitting of concrete pieces from the heated surfaces, greatly jeopardizing the load carrying capacity and durability. Spalling is closely correlated with the evolution and distribution of pore-pressure $p^g$ and temperature $T$ in heated concrete. Conventionally complicated thermo-hydro-chemical (THC) models are necessary for capturing this information. In this work, we proposed a hybrid neural network for quickly obtaining $p^g$, $T$ of heated concrete. The neural network includes two parts: (i) a well-established autoencoder (AE) and (ii) a fully connected neural network (FNN). A strongly coupled THC model was first used to provide large amounts of results represented by thousands RGB images. The AE was used to condense the images into characteristic vectors, which were used for training the FNN. After training, the FNN can be used for predicting the corresponding characteristic vectors considering different concrete properties, moisture and fire loadings. Then the decoder of the AE is used to translate the characteristic vectors into RGB images, storing the information of $p^g$ and $T$. Numerical tests indicate the effectiveness and reliability of the proposed model.

Active Exoskeletons can become a powerful tool for therapists for the rehabilitation of patients suffering from neurophysiological conditions. The mathematical modeling for estimating joint moments required for human walking movement proves difficult due to the high number of degrees of freedom (DoF) and the complexity of movement. Another factor that poses a problem is the unavailability of ground reaction force (GRF) data, which must be present as the external applied forces in the model. This paper presents a machine learning-based approach for predicting joint moments for walking that uses only the kinematic data of the subjects. The dataset used includes data available from published sources as well as data collected by the authors. The predictions have been compared with and validated using the joint moment results from optimization-based inverse dynamics model in OpenSim. Subsequently, a concept design of a lower limb exoskeleton has been presented and actuator requirements for the same are set according to the joint moment predictions for a specific human subject. The prototype design includes eight rotational degrees of freedom (DOF) in total, i.e., four degrees of freedom per leg: two at the hip joint, one at the knee joint and one at the ankle joint. The feasibility study of the prototype has been carried out with the help of finite element analysis (FEA) in Ansys software after utilizing the weight of the human being and joint rotations as inputs to the model. Based on the results obtained from the FEM, the design has been optimized to ensure structural stability.

This paper describes a neural network (NN)-based topology optimization approach for designing microstructures. The design variables are the NN weights and biases used to describe the density field, which is independent of element meshes. The number of design variables and gray elements is reduced substantially, and no filtering is necessary. Three numerical examples are provided to demonstrate the efficacy of the proposed method, namely, maximum shear modulus, maximum bulk modulus, and negative Poisson’s ratio.

In this paper, we propose a new computation scheme for numerical solutions of Kolmogorov equations based on a high-order weak approximation method of stochastic differential equations and deep learning. The scheme provides a spatial approximation for solving Kolmogorov equations without the curse of dimensionality. We show numerical examples based on the proposed scheme for high-dimensional Kolmogorov equations.

Collapses are the most sensational types of events and frequently the ones that cause the most serious consequences in tunneling operations. It often occurs because of insufficient geological studies and the limitations of experience-based decision-making. To cope with those problems, we proposed a data-driven model based on the tunnel boring machine operational data and the in situ geological information to forecast tunneling-induced ground collapse. In the proposed model, we offered a general data process flow diagram to process engineering data. Three machine learning classifiers, k-nearest neighbors, support vector classifier, and random forests were adopted for collapse prediction. The performance of the three classifiers was verified based on the data from the Yinsong water conveyance tunnel. The results illustrated that the proposed data-driven model was sufficient for the studying task with 90% of the collapsed zones being identified on average. The contributions of this paper are to provide a reliable data process flow diagram to process engineering data and offer an accurate and robust model for identifying collapses.

In this paper, we present a deep energy method for functionally graded beams based on both Euler–Bernoulli and Timoshenko beam theory to study their mechanical and thermal properties. We consider the effect of temperature as well as porosity under saturated and unsaturated conditions. The objective function is related to the total potential energy (and boundary conditions) and minimized through neural network training. The results are validated by comparison with benchmark problems available in the literature.

Adhesively bonded joints showcase non-uniform stress distribution, along their length as the load is transferred through layers of dissimilar stiffness. For efficient transfer of loads, the peak interfacial shear stress is required to be engineered. In this study, inspired by electric pulses, the interphase modulus is modified according to square, sinusoidal and triangular pulses. The variation in peak stresses with increased number of pulses up to four is also investigated. The developed analytical model is solved for the interfacial shear stresses as well as the peel stresses, using energy functional approach, through MAPLE software. The abrupt changes in modulus in square pulse graded interphase are observed to create highest interfacial shear stresses among the considered grading profiles. Furthermore, the peak interfacial stresses are observed to increase with increased number of pulses. An effective elastic modulus parameter is defined to indicate the area under the modulus profile curve. The effective modulus is found to be gradually increasing with increase number of pulses in square graded interphase. Whereas, it is constant for sinusoidal- and triangular-graded interphases. A deep machine learning-based physics informed neural network model is developed to quickly solve the developed governing differential equations. Therefore, results from the machine leaning model are compared to the analytical results.