Inverse Neural Network Research Articles

Inversion of the 3D in situ stress field in a complex valley area around the underground cavern group of the Yingliangbao Hydropower Station

The inversion of three-dimensional in situ stress fields in complex river valleys is of great significance for evaluating the excavation stability of deep underground caverns. Based on known in situ stress measurements, it is a major method for obtaining the geostress of the engineering area through the inversion geostress of the whole area with the application of geo-mathematical mechanics. However, the inversion results are often limited by the geological structure and heterogeneous alterated rock, as well as the measurement data reliability. A three-dimensional nonlinear neural network inversion was combined with a simulation of stratigraphic denudation by stages, tectonic compression, and alteration degree in the fault structure. The model was then applied to an underground cavern group of the Yingliangbao Hydropower Station. Comparative analysis showed that the inversion results were in reasonable accordance with the measured in situ stress values. Moreover, the reliability of the three-dimensional in situ stress field obtained using this method was further verified from the stress-induced failure during the cavern excavation. This technique is helpful for optimizing the design of subsequent excavation and reduces the potential instability of surrounding rock during a layered excavation.

Combining autoencoder neural network and Bayesian inversion to estimate heterogeneous permeability distributions in enhanced geothermal reservoir: Model development and verification

Determining permeability distributions in reservoirs is critical for the management of limited earth resources. While hydraulic fracturing is widely used to enhance the permeability of deep geothermal, gas and oil reservoirs, it remains challenging to infer heterogeneous distributions of permeability. Typically, a limited number of boreholes are available at which reservoir imaging and tracer testing can be conducted. The number of observations is often far fewer than the number of estimable permeability parameters, making model inversion ill-posed. To overcome this problem, the autoencoder neural network was combined with a Bayesian inversion algorithm based on Markov Chain Monte Carlo (MCMC) sampling, in order to estimate the spatial distributions of permeability in an enhanced geothermal reservoir, conditional to temperature and outflow rate observations from a single-well-injection-withdrawal test (SWIW). The autoencoder neural network was used to reduce parameter dimensionality by four orders of magnitude. MCMC sampling was used to estimate low-dimensional parameters via inversion of SWIW observations. A high-resolution permeability distribution was reconstructed from the low-dimensional parameterization through reapplication of the autoencoder neural network. Application to a synthetic enhanced geothermal system demonstrated that the methodology achieved rapid stabilization and low permeability estimation error (<10%). By combining deep-learning method with Bayesian inversion, permeability distributions in geo-energy reservoirs can be estimated from a limited set of borehole data.

A Novel Constitutive Parameters Identification Procedure for Hyperelastic Skeletal Muscles Using Two-Way Neural Networks

Muscle soreness can occur after working beyond the habitual load, especially for people engaged in high-intensity work load. Prediction of hyperelastic material parameters is essentially an inverse process, which possesses challenges. This work presents a novel procedure that combines nonlinear finite element method (FEM), two-way neural networks (NNs) together with experiments, to predict the hyperelastic material parameters of skeletal muscles. FEM models are first established to simulate nonlinear deformation of skeletal muscles subject to compressions. A dataset of nonlinear relationship between nominal stress and principal stretch of skeletal muscles is created using our FEM models. The dataset is then used to establish two-way NNs, in which a forward NN is trained and it is in turn used to train the inverse NN. The inverse NN is used to predict the hyperelastic material parameters of skeletal muscles. Finally, experiments are carried out using fresh skeletal muscles to validate the predictions in great detail. In order to examine the accuracy of the two-way NNs predicted values against the experimental ones, a decision coefficient [Formula: see text] with penalty factor is introduced to evaluate the performance. Studies have also been conducted to compare the present two-way NNs approach with the other existing methods, including the directly (one-way) inverse problem NN, and improved niche genetic algorithm (INGA). The comparison results show that two-way NNs model is an accurate approach to identify the hyperelastic parameters of skeletal muscles. The present two-way NNs method can be further expanded to the predictions of constitutive parameters of other type of nonlinear materials.

Convolutional neural network inversion of airborne transient electromagnetic data

ABSTRACTAs an efficient geophysical exploration technique, airborne transient electromagnetics shows strong adaptability to complex terrains and can provide subsurface resistivity information rapidly with a dense spatial coverage. However, the huge volume of airborne transient electromagnetic data obtained from a large number of spatial locations presents a great challenge to real‐time airborne transient electromagnetic interpretation due to the high computational cost. Moreover, the inherent non‐uniqueness of the inverse problem also limits our ability to constrain the underground resistivity structure. In this study, we develop an entirely data‐driven convolutional neural network to solve the airborne transient electromagnetic inverse problem. Synthetic tests show that the convolutional neural network is computationally efficient and yields robust results. Compared with the Gauss–Newton method, convolutional neural network inversion does not depend on the choices of an initial model and the regularization parameters and is less prone to getting trapped in a local minimum. We also demonstrate the general applicability of the convolutional neural network to three‐dimensional synthetic airborne transient electromagnetic responses and the field observations acquired from Leach Lake Basin, Fort Irwin, California. The efficient convolutional neural network inversion framework can support real‐time resistivity imaging of subsurface structures from airborne transient electromagnetic observations, providing a powerful tool for field explorations.

Inverse neural networks modelling of a piezoelectric stage with dominant variable

Inverse neural networks modelling of a piezoelectric stage with dominant variable

Optimal Design of Earthquake-Resistant Buildings Based on Neural Network Inversion

An effective seismic design entails many issues related to the capacity-based assessment of the non-linear structural response under strong earthquakes. While very powerful structural calculation programs are available to assist the designer in the code-based seismic analysis, an optimal choice of the design parameters leading to the best performance at the lowest cost is not always assured. The present paper proposes a procedure to cost-effectively design earthquake-resistant buildings, which is based on the inversion of an artificial neural network and on an optimization algorithm for the minimum total cost under building code constraints. An exemplificative application of the method to a reinforced-concrete multi-story building, with seismic demands corresponding to a medium-seismicity Italian zone, is shown. Three design-governing parameters are assumed to build the input matrix, while eight capacity-design target requirements are assigned for the output dataset. A non-linear three-dimensional concentrated plasticity model of the structure is implemented, and time-history dynamic analyses are carried out with spectrum-consistent ground motions. The results show the promising ability of the proposed approach for the optimal design of earthquake-resistant structures.

A hybrid residual neural network–Monte Carlo approach to invert surface wave dispersion data

ABSTRACTSurface‐wave inversion is a non‐linear and ill‐conditioned problem usually solved through deterministic or global optimization approaches. Here, we present an alternative method based on machine learning. Under the assumption of a local one‐dimensional model, we train a residual neural network to predict the non‐linear mapping between the full dispersion image and the model space, parameterized in terms of shear wave velocity and layer thicknesses. On the one hand, compared to standard convolutional neural networks, the residual network prevents the vanishing gradient problem when training a deep network. On the other hand, the use of the full dispersion image avoids the time‐consuming and often ambiguous picking procedure and allows considering higher modes in the inversion framework. One key aspect of any machine learning inversion strategy is the definition of an appropriate training set. In this case, the models forming the training and validation examples are uniformly drawn from previously defined ranges that cover a wide range of possible near‐surface layered Vs models. The reflectivity method constitutes the forward modelling operator that converts the model parameters into the observed shot gathers. The inversion also includes a Monte Carlo simulation strategy that propagates onto the model space the uncertainties related to noise in the data and the modelling error introduced by the network approximation. We first discuss synthetic inversions to assess the applicability of the proposed method and to analyse the effect of erroneous model parameterizations. The inversion results are also benchmarked with those provided by a more standard approach in which the particle swarm optimization algorithm inverts the fundamental mode only. Then, we discuss a field data application. Our tests confirm that the residual neural network inversion provides accurate model estimations and reliable uncertainty appraisals. One of the main benefits of the proposed approach is that once the network is trained it provides the near‐surface shear wave velocity profile in near real‐time.

Development and Application of a Multi-Objective Tool for Thermal Design of Heat Exchangers Using Neural Networks

This paper presents the design of a multi-objective tool for sizing shell and tube heat exchangers (STHX), developed under a University/Industry collaboration. This work aims to show the feasibility of implementing artificial intelligence tools during the design of Heat Exchangers in industry. The design of STHX optimisation tools using artificial intelligence algorithms is a visited topic in the literature, nevertheless, the degree of implementation of this concept is uncommon in industrial companies. Thus, the challenge of this research consists of the development of a tool for the design of STHX using artificial intelligence algorithms that can be used by industrial companies. The approach is implemented using a simulated dataset contrasted with ARA TT, the company taking part in the project. The given dataset to develop a theoretical STHX calculator was modeled using MATLAB. This dataset was used to train seven neural networks (NNs). Three of them were mono-objective, one per objective to predict, and four were multi-objective. The last multi-objective NN was used to develop an inverse neural network (INN), which is used to find the optimal configuration of the STHXs. In this specific case, three design parameters, the pressure drop on the shell side, the pressure drop on the tube side and heat transfer rate, were jointly and successfully optimised. As a conclusion, this work proves that the developed tool is valid in both terms of effectiveness and user-friendliness for companies like ARA TT to improve their business activity.

A Neural Network-Based Hybrid Framework for Least-Squares Inversion of Transient Electromagnetic Data

Inversion of large-scale time-domain transient electromagnetic (TEM) surveys is computationally expensive and time-consuming. The calculation of partial derivatives for the Jacobian matrix is by far the most computationally intensive task, as this requires calculation of a significant number of forward responses. We propose to accelerate the inversion process by predicting partial derivatives using an artificial neural network. Network training data for resistivity models for a broad range of geological settings are generated by computing partial derivatives as symmetric differences between two forward responses. Given that certain applications have larger tolerances for modeling inaccuracy and varying degrees of flexibility throughout the different phases of interpretation, we present four inversion schemes that provide a tunable balance between computational time and inversion accuracy when modeling TEM datasets. We improve speed and maintain accuracy with a hybrid framework, where the neural network derivatives are used initially and switched to full numerical derivatives in the final iterations. We also present a full neural network solution where neural network forward and derivatives are used throughout the inversion. In a least-squares inversion framework, a speedup factor exceeding 70 is obtained on the calculation of derivatives, and the inversion process is expedited ~36 times when the full neural network solution is used. Field examples show that the full nonlinear inversion and the hybrid approach gives identical results, whereas the full neural network inversion results in higher deviation but provides a reasonable indication about the overall subsurface geology.

A Multiprocessing Scheme for PET Image Pre-Screening, Noise Reduction, Segmentation and Lesion Partitioning.

Accurate segmentation and partitioning of lesions in PET images provide computer-aided procedures and doctors with parameters for tumour diagnosis, staging and prognosis. Currently, PET segmentation and lesion partitioning are manually measured by radiologists, which is time consuming and laborious, and tedious manual procedures might lead to inaccurate measurement results. Therefore, we designed a new automatic multiprocessing scheme for PET image pre-screening, noise reduction, segmentation and lesion partitioning in this study. PET image pre-screening can reduce the time cost of noise reduction, segmentation and lesion partitioning methods, and denoising can enhance both quantitative metrics and visual quality for better segmentation accuracy. For pre-screening, we propose a new differential activation filter (DAF) to screen the lesion images from whole-body scanning. For noise reduction, neural network inverse (NN inverse) as the inverse transformation of generalized Anscombe transformation (GAT), which does not depend on the distribution of residual noise, was presented to improve the SNR of images. For segmentation and lesion partitioning, definition density peak clustering (DDPC) was proposed to realize instance segmentation of lesion and normal tissue with unsupervised images, which helped reduce the cost of density calculation and completely deleted the cluster halo. The experimental results of clinical data demonstrate that our proposed methods have good results and better performance in noise reduction, segmentation and lesion partitioning compared with state-of-the-art methods.

Using Generalized Regression Neural Network to Retrieve Bare Surface Soil Moisture From Radarsat-2 Backscatter Observations, Regardless of Roughness Effect

The combined influence of surface soil moisture and roughness on radar backscatters has been limiting SAR’s application in soil moisture retrieval. In the past research, multi-temporal analysis and artificial neural network (ANN) inversion of physically based forward models were regarded as promising methods to decouple that combined influence. However, the former does not consider soil roughness change over a relatively longer period and the latter makes it hard to thoroughly eliminate the effect of soil roughness. This study proposes to use generalized regression neural network (GRNN) to derive bare surface soil moisture (BSSM) from radar backscatter observations regardless of the effect of soil roughness (GRNN inversion of backscatter observations). This method not only can derive BSSM from radar backscatters, provided soil roughness is unknown in any long period, but also can train models based on small-size sample data so as to reduce the manual error of training data created by simulation of physically based models. The comparison of validations between BSSM-backscatter models and BSSM-roughness-backscatter models both analyzed by GRNN shows that the incorporation of soil roughness cannot raise the prediction accuracy of models and, instead, even reduce it, indicating that the combined influence is thoroughly decoupled when being analyzed by GRNN. Moreover, BSSM-backscatter models by GRNN are recommended due to their good prediction, even compared to those related models in past publications.

A deep neural network inverse solution to recover pre-crash impact data of car collisions

In this work, we have successfully developed a data-driven artificial intelligence (AI) inverse problem solution for traffic collision reconstruction. In specific, we have developed and implemented a machine learning computational algorithm and built a deep neural network to determine and identify the initial impact conditions of car crash based on its final material damage state and permanently deformed structure configuration (wreckage).In this work, we have demonstrated that the developed machine learning algorithm as an inverse problem solver can accurately identify initial collision conditions in an inverse manner, which are practically unique if we use permanent plastic deformation as the forensic data signatures. In other words, we think that the massive plastic energy dissipation process and the related big data will make final structure damage state insensitive to the initial car collision conditions. Thus, it provides an inverse solution for car crash forensic analysis by reconstructing the initial failure load parameters and conditions based on the permanent plastic deformation distribution of cars. This approach has general significance in solving the inverse problem for engineering failure analysis and vehicle crashworthiness analysis, which provides a key contribution for the unmanned autonomous vehicle and the related technology.

Neural Network Inversion-Based Model for Predicting an Optimal Hardware Configuration

Deciding the number of processors that can efficiently speed-up solving a computationally intensive problem while perceiving efficient power consumption constitutes a major challenge to researcher in the HPC high performance computing realm. This paper exploits machine learning techniques to propose and implement a recommender system that recommends the optimal HPC architecture given the problem size. An approach for multi-objective function optimization based on neural network (neural network inversion) is employed. The neural network inversion approach is used for forward problem optimization. The objective functions in concern are maximizing the speedup and minimizing the power consumption. The recommendations of the proposed prediction systems achieved more than 89% accuracy for both validation and testing set. The experiments were conducted on 2500 CUDA core on Tesla K20 Kepler GPU Accelerator and Intel(R) Xeon(R) CPU E5-2695 v2.

Magnetic Properties Identification by Using a Bi-Objective Optimal Multi-Fidelity Neural Network

In order to identify the magnetic properties of magnetic steel, the synergy between the data arising from the experimental activity, an FE model, and the use of a multi-fidelity surrogate could relieve the burden of the total cost. A neural network, with as many outputs as fidelity levels, is adopted in quality of metamodel to describe the forward problem [forward neural network (FNN)]. FNN is trained using multiple losses aiming at getting a robust surrogate that is poorly sensitive to the chosen norm. This makes it bi-objective optimal since several error metrics are simultaneously minimized. In addition, a conjugate, inverse net (INN <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">CJ</sub> ) is built, which is a ready-to-use tool for inverse properties identification, since no optimization runs are required. Its performances are compared to those obtained with a transfer learning-based approach (INN <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">TR</sub> ) and a single-fidelity inverse neural network (INN <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SF</sub> ). Finally, a real B - H curve identification task has been solved, thereby validating the conjugate inverse net.

Research on Inversion Mechanism of Chlorophyll—A Concentration in Water Bodies Using a Convolutional Neural Network Model

For Case-II water bodies with relatively complex water qualities, it is challenging to establish a chlorophyll-a concentration (Chl-a concentration) inversion model with strong applicability and high accuracy. Convolutional Neural Network (CNN) shows excellent performance in image target recognition and natural language processing. However, there little research exists on the inversion of Chl-a concentration in water using convolutional neural networks. Taking China’s Dongting Lake as an example, 90 water samples and their spectra were collected in this study. Using eight combinations as independent variables and Chl-a concentration as the dependent variable, a CNN model was constructed to invert Chl-a concentration. The results showed that: (1) The CNN model of the original spectrum has a worse inversion effect than the CNN model of the preprocessed spectrum. The determination coefficient (RP2) of the predicted sample is increased from 0.79 to 0.88, and the root mean square error (RMSEP) of the predicted sample is reduced from 0.61 to 0.49, indicating that preprocessing can significantly improve the inversion effect of the model.; (2) among the combined models, the CNN model with Baseline1_SC (strong correlation factor of 500–750 nm baseline) has the best effect, with RP2 reaching 0.90 and RMSEP only 0.45. The average inversion effect of the eight CNN models is better. The average RP2 reaches 0.86 and the RMSEP is only 0.52, indicating the feasibility of applying CNN to Chl-a concentration inversion modeling; (3) the performance of the CNN model (Baseline1_SC (RP2 = 0.90, RMSEP = 0.45)) was far better than the traditional model of the same combination, i.e., the linear regression model (RP2 = 0.61, RMSEP = 0.72) and partial least squares regression model (Baseline1_SC (RP2 = 0.58. RMSEP = 0.95)), indicating the superiority of the convolutional neural network inversion modeling of water body Chl-a concentration.

Analysis of Surrounding Rock Creep Effect on the Long‐Term Stability of Tunnel Secondary Lining

The secondary lining failure of deep buried soft rock tunnel often occurs, which is obviously related to the time factor. The formation mechanism of this phenomenon is studied in this paper. Therefore, the combination of in situ stress measurement and neural network inversion is used to analyze the distribution characteristics of in situ stress. At the same time, the creep characteristics of surrounding rock are tested in laboratory, and the key parameters are obtained. Combined with the characteristics of surrounding rock, the calculation model is established by using discrete element simulation technology and considering the joints of surrounding rock. According to the above multiple information, the stress characteristics of the secondary lining in different time periods are analyzed creatively. Finally, the method of setting arch and adding anchor bolt in key parts is proposed, and significant effect results are obtained.

Inversion of Temperature and Humidity Profile of Microwave Radiometer Based on BP Network

In this paper, the inversion method of atmospheric temperature and humidity profiles via ground-based microwave radiometer is studied. Using the three-layer BP neural network inversion algorithm, four BP neural network models (temperature and humidity models with and without cloud information) are established using L-band radiosonde data obtained from the Atmospheric Exploration base of the China Meteorological Administration from July 2018 to June 2019. Microwave radiometer level 1 data and cloud radar data from July to September 2019 are used to evaluate the model. The four models are compared with the measured sounding data, and the inversion accuracy and the influence of cloud information on the inversion are subsequently analyzed. The results show the following: the average errors of temperature and humidity profiles for the model without cloud information are 1.18°C and 11.7%, while the average errors of temperature and humidity profiles for the model with cloud information are 0.71°C and 6.09%. Compared with the profiles that lack cloud information, the RMSE of most altitudes is reduced to some extent after cloud information is added, which is particularly obvious at layers where cloud is present.

Wind speed and power prediction using MK-PINN

A multi-kernel pseudo inverse neural network (MKPINN) is proposed in this paper for efficient wind speed and power forecasting. The proposed model has been compared with Gaussian, wavelet, polynomial and sigmoid kernel. To get best output and learning methodology and stability pseudo inverse neural network is added, which substitutes the hidden layer with kernel function. This helps to achieve more accurate and faster response. Various case studies have been carried out from ten minutes to five hours interval in order to prove it accuracy.

Application of laser profilo metry in problems of welding equipment geometric adaptation

Adaptive adjustment of the relationship between the welding process parameters and the butt geometry permits to reduce the likelihood of welding defects appearance and improve the quality of the welded joint in automatic welding of large-diameter pipes. To obtain data on the configuration of the welded joint, the RF627 laser vision sensor is used. To reduce the influence of restrictions arising during the welding process, a median algorithm for filtering impulse noise is proposed. To calculate the geometric parameters of the welded joint, a model based on pixel data obtained from a laser sensor is proposed. The restoration of the welded butt parameters is carried out according to the algorithm of piecewise-linear approximation, which involves the determination of six characteristic points of the butt. The adaptive adjuster uses an inverse neural network model for adjustment the parameters of the welding process: welding current, voltage, wire feed speed. To train the neural network, the characteristic parameters of the welded butt are used: gap, skewing (warping of the edges) and bluntness (for the root weld), the current width of the butt groove in each layer (for other types of welds). The weights of the neural network layers are restored online using a gradient descent algorithm. The important role of the laser vision sensor in solving the problem of adaptation of welding equipment and the effectiveness of the proposed algorithms are confirmed experimentally. Keywords laser vision sensor; robotic welding; multilayer/multi-pass welding; piecewise linear approximation; adaptive control with a reverse neural network model

Disturbance rejection based on iterative learning control with extended state observer for a four-degree-of-freedom hybrid magnetic bearing system

Iterative learning control (ILC) is an iterative control strategy which calculates a new input according to the error in previous cycles. It is widely used in industries with repetitive operations. Since magnetic bearing systems can be considered as running in a repetitive task, the ILC could be applied. Hence, this paper proposes a modified iterative learning control (ILC) strategy for a four degree-of-freedom (DOF) hybrid magnetic bearing system to reject the disturbance based on an extended state observer (ESO). The feasibility of applying the ILC to the four-DOF magnetic bearing system is analyzed firstly. It is proved that the tracking error can converge by selecting suitable controller parameters. Secondly, to accurately obtain the uncertainties in the operation process, an ESO is designed. As for a repetitive rotation process, the iteration variant disturbance may have a certain influence on the performance of the system which is not usually considered. Therefore, the iteration variant disturbance is introduced and the effectiveness is derived. Finally, simulations and experiments are carried out to demonstrate the effectiveness of the proposed method. The classical proportion-integration-differentiation (PID) control and an existing proposed neural network inverse (NNI) control are implemented for comparison. The results show that the proposed strategy can achieve better reference tracking and disturbance suppression ability than PID and NNI control.