Design optimization of adaptive systems requires a robust analysis method that can accommodate various changes in design and boundary conditions. In this work, physics-informed neural networks (PINNs) are used to approximate solutions to differential equations across a range of problem parameter values. This mesh-free method simply requires residual evaluation at sampling points within the analysis domain and along boundaries, and the training process does not require any reference problem to be solved through conventional solution methods. The trained model can be used to predict the solution field, conduct parameter space analysis and design optimization. Using automatic differentiation, the design objective and their derivatives can be computed as a post process for a gradient-based design optimization. The method is demonstrated in a 1D heat transfer problem governed by the steady-state heat equation. Use of the PINN model for design optimization is illustrated in a problem of finding a material transition location to minimize temperature at a specified location. The PINN model that does not include problem parameters as input can be trained to within 0.05% error. PINN models that involve problem parameters as inputs are more difficult to train, especially when the input-to-output relationship is complex.

Predictive Maintenance (PMx) methods leverage available heritage data, maintenance records, and vehicle information to forecast rotorcraft failure modes before they become a problem. The goal of PMx is to decrease cost and vehicle downtime while increasing availability. When the required data are incomplete or corrupted, the worst case (grossly conservative) scenario must be assumed and unnecessary costs are incurred. In this manuscript we propose data-science methods to identify and characterize regions of data corruption, and machine-learning (ML) techniques to address the problem of missing and corrupted tri-axial H-60 4G accelerometer data for PMx for a H-60 rotorcraft. Accurate 4G sensor readings are a critical component of helicopter flight regime recognition and flight damage assessment. In contrast to the traditional time-series prediction approach, which commonly use recurrent or long short-term memory (LSTM) networks, our proposed methods use a simpler deep neural network (DNN) to reconstruct the 4G accelerometer signal independently at every time instant. We demonstrate that the DNN approach is a viable option for sensor reconstruction independent of the length of period the sensor malfunctions.

With the rapid development of rail traffic, the importance of railway overhaul is becoming increasingly prominent. Making an inventory on tools is an important step that railway workers must take before and after railway inspection. The tools left on the railway will cause great harm to train safety. To avoid this happening, the commonly used method is manual inventory at present, which is time-consuming, laborious and easily leads to omissions. In order to overcome these shortcomings, this paper proposes a Faster Region-based Convolutional Neural Network (Faster R-CNN)-based method for tools inventory. To realize the method, a Faster R-CNN architecture based on ZF-Net is modified and a database including a large number of images for 10 types of tools is built. Then the Faster R-CNN is trained and validated using the built database. The performance of the trained Faster R-CNN is evaluated using some new images which are not be used for training process. The result shows 95.0325% average precision (AP) ratings for 10 different types of tools and proves the proposed method is effective.

In the new data intensive world, predictive maintenance has become a central issue for the modern industrial plants. Monitoring of electric machinery is one of the most important challenges in predictive maintenance. Adaptive manufacturing processes/plants may be possible through the monitored conditions. In this respect, several attempts have been made to utilize deep learning algorithms for rotating machinery fault detection and diagnosis. Among them, deep autoencoders are very popular, because of their denoising effect. They are also implemented in electric machinery fault diagnostics in order to obtain lower order representation of signals. However, none of these efforts regard the autoencoders as compression units. Bearing in mind that spectra of vibration and current signals that are collected from electric machinery are critical instruments for detection and diagnosis of their faults, we propose that deep stacked autoencoder can be utilized as spectrum compression units. The performance of the proposed strategy are assessed using a bearing data set in three ways: (1)Rule-based classifiers are implemented on raw and compressed-decompressed spectrum and their performance are compared. (2) It is shown that the several machine learning classifiers such as support vector machines, artificial neural networks and k-nearest neighbour classifiers on compressed-decompressed spectrum achieves the performance of them on raw data. (3) A multi-layer perceptron (MLP) classifier is implemented on the low dimensional representation and it is demonstrated that the strategy of employing the same autoencoder as pretraining of feature extraction module cannot outperform the performance of this MLP classifier.

Bridge management and maintenance work is an important part for the assessment the health state of bridge. The conventional management and maintenance work mainly relied on experienced engineering staffs by visual inspection and filling in survey forms. However, the human-based visual inspection is a difficult and time-consuming task and its detection results significantly rely on subjective judgement of human inspectors. To address the drawbacks of human-based visual inspection method, this paper proposes an image-based comprehensive maintenance and inspection method for bridges using deep learning. To classify the types of bridges, a convolutional neural network (CNN) classifier established by fine-turning the AlexNet is trained, validated and tested using 3832 images with three types of bridges (arch, suspension and cable-stayed bridge). For the recognition of bridge components (tower and deck of bridges), a Faster Region-based Convolutional Neural Network (Faster R-CNN) based on modified ZF-net is trained, validated and tested by utilizing 600 bridge images. To implement the strategy of a sliding window technique for the crack detection, another CNN from fine-turning the GoogLeNet is trained, validated and tested by employing a databank with cropping 1455 raw concrete images into 60000 intact and cracked images. The performance of the trained CNNs and Faster R-CNN is tested on some new images which are not used for training and validation processes. The test results substantiate the proposed method can indeed recognize the types and components and detect cracks for a bridges.

High-speed railway plays critical roles in public safety and the country’s economy. Visual detection of components and damages can reflect the health conditions of high-speed railway. Human-based visual inspection is a difficult and time-consuming task and its detection results significantly rely on subjective judgement of human inspectors. Image-based detection methods abandon the weakness of human-based visual inspection. However, in practice, the complex real-world situations, such as lighting and shadow changes, can lead to challenges to the wide adaptability of image process techniques. To overcome these challenges, this paper provides a Faster Region-based Convolutional Neural Network (Faster R-CNN)-based detection method of component types and track damage for high-speed railway. To realize the method, a database including 575 images labeled for three component types and one track damage type of high-speed railway is built. A Faster R-CNN architecture based on ZF-Net is modified, then trained and validated using the built database. The performance of the trained Faster R-CNN is evaluated using 50 new images which are not be used for training process. The results show that the proposed method can indeed detect the component types and track damage for high-speed railway.

In the last couple of years, advancements in the deep learning, especially in convolutional neural networks, proved to be a boon for the image classification and recognition tasks. One of the important practical applications of object detection and image classification can be for security enhancement. If dangerous objects or scenes can be identified automatically, then a lot of accidents can be prevented. For this purpose, in this paper we made use of state-of-the-art implementation of Faster Region-based Convolutional Neural Network (Faster R-CNN) based on the monitoring video of hoisting sites to train a model to detect the dangerous object and the worker. By extracting the locations of them, object-human interactions during hoisting, mainly for changes in their spatial location relationship, can be understood whereby estimating whether the scene is safe or dangerous. Experimental results showed that the pre-trained model achieved good performance with a high mean average precision of 97.66% on object detection and the proposed method fulfilled the goal of dangerous scenes recognition perfectly.

Engineering systems subject to high-rate extreme environments can often experience a sudden plastic deformation during a dynamic event. Examples of such systems include civil structures exposed to blast or aerial vehicles experiencing impacts. The change in configuration through deformation can rapidly lead to catastrophic failures resulting in intolerable losses in investments or human lives. A solution is to conduct fast system estimation enabling real-time decisions, in the order of microseconds, to mitigate such high-rate changes. To do so, we propose a model-driven observer coupled with a data-driven adaptive wavelet neural network to provide real-time stiffness estimations to continuously update a system’s model. This real-time system identification method offers adaptability of the system’s parameters to unforeseeable changes. The results of the simulations demonstrate accurate stiffness estimations in milliseconds for three different excitation conditions for a one degree-of-freedom spring, mass, and damper system with variable stiffness.

Due to the particularity of texture features in ancient buildings, which refers to the fact that these features have a high historical and artistic value, it is of great significance to identify and count them. However, the complexity and large number of textures are a big challenge for the artificial identification statistics. In order to overcome these challenges, this paper proposes an approach that uses smartphones to achieve a real-time detection of ancient buildings’ features. The training process is based on SSD-Mobilenet, which is a kind of Convolutional Neural Network (CNN). The results show that this method shows well performance in reality and can indeed detect different ancient building features in real time.

The technology of swarm intelligence has been applied to a mechanical vibration monitoring system composed of a network of units equipped with sensors and actuators. The expression of “swarm intelligence” was first used in 1988 in the context of cellular robotic systems, where lots of simple agents may generate self-organized patterns through mutual interactions. There are various examples of the swarm intelligence in the natural environment, a swarm of ants, birds or fish. In this sense, the network of agents in a swarm may have some kind of intelligence or higher function than those appeared in a simple agent, which is defined as the swarm intelligence. The concept of swarm intelligence may be applied in diverse engineering fields such as flexible pattern recognition, adaptive control system, or intelligent monitoring system, because some kind of intelligence may emerge on the network without any special control system. In this study, a simulation model of a five degree-of-freedom lumped mass-spring system was prepared as an example of a mechanical dynamic system. Five units composed of a displacement sensor and a variable damper as actuator were assumed to be placed on each mass of the system. Each unit was connected to each other to exchange the information of state variables measured by sensors on each unit. Because the network of units configured as a mutual connected neural network, a kind of artificial intelligence, the network of units may memorize the several expected vibration-controlled patterns and may produce the signal to the actuators on the unit to reduce the vibration of target system. The simulation results showed that the excited vibration was reduced autonomously by selecting the position where the damping should be applied.

Fly by feel is a concept in which distributed sensors and actuators are integrated on an aerial system for state awareness or sensation of the environment, and make use of distributed control to increase the system maneuverability, stability and safety. Artificial hair sensors are good candidates as sensors for the fly by feel concept because they are lightweight, have low manufacturing costs and can easily be integrated on the surface of air-vehicle without affecting the flow. We investigate an application of artificial hair sensors considering its capability of measuring the local flow velocity combined with a Feedforward Artificial Neural Network to predict the aerodynamic quantities such as lift coefficient, moment coefficient, angle of attack and free-stream velocity in real-time. These quantities, when combined with the physical and unsteady aerodynamics parameters, will make a framework for designing and implementing an active controller for gust alleviation in a pitch and plunge airfoil system.

Oil and gas infrastructures may be exposed to landslides, earthquakes, corrosion and fatigue, and to damage from thefts or vandalism, leading to leakage and failure with serious economic and ecologic consequences. For this reason, an increasing interest in applied research on monitoring and protecting pipelines (for fuel, oil and natural gas transportation) arises. Aimed at the mitigation of catastrophic effects of human and natural damage, the present paper proposes a smart real-time Structural Health Monitoring (SHM) system capable to control structural integrity continuously, focusing on the issue of spillage for thefts of fuels which are not detectable, in real-time, by the existing monitoring systems. The system consists of a smart-pipeline containing a health monitoring integrated measurement chain, i.e. an enhanced Fiber Bragg Gratings-based fiber optics neural network on the pipes, for displacement and acceleration monitoring (gathering many other different measurements such as: ground motion, permanent ground displacement, pipeline temperature, pipeline deformation, leakage, etc.). Specifically, the ability to measure these characteristics at hundreds of points along a single fiber and the great accuracy of each point of measure, are particularly interesting for the monitoring of structures such as pipelines in order to detect hazardous and unauthorized intrusion and damage.

Shape memory alloys (SMA) when subjected to deformation at low temperature can recover their original shape by heating above a temperature called Austenite transformation temperature. This original shape is sustained till the material is deformed again by an applied stress. This property makes the SMA a unique actuator, which doesn’t require any other components. Also, the material’s resistance changes with deformation. Thus the change in resistance can be used to sense the deformation, which eliminates the requirements of additional sensors. This can make the system more compact and reduce the cost. In our study, a binary Nickel-Titanium alloy is used as a rotary actuator. The actuation is controlled by adjusting the temperature through controlled joule heating by varying the electric current. The manipulator used in this research is a single degree-of-freedom, bias type actuator. SMA actuation in this system is under a varying stress, thereby creating a complex thermo-mechanical condition which affects the transformation temperatures, significantly. Also, the resistance change during heating and cooling paths exhibit hysteresis behavior. This paper investigates the use of artificial neural network (ANN) in establishing relationship between resistance and angular position of the manipulator. To model the hysteresis behavior of the SMA, in addition to resistance of the SMA, other electric properties like voltage etc are given as input to the ANN. The obtained ANN model is able to determine the angular position of the rotary manipulator with good accuracy.

In this study spatially shaded PVDF was incorporated into a compression sleeve and used to create a wearable joystick which was able to identify and classify gestures made by the right hand. A multilayered feedforward neural network was used to discriminate movements of the hand at the wrist. In feedforward operation, the output voltage of the PVDF was collected using a DAQ system and used to populate an updating input vector which was fed into the network for real-time pattern classification. The network was trained using traditional backpropagation methods where the training inputs were an assortment of collected and simulated voltage patterns of three specific hand gestures and the outputs were specified vectors corresponding to said hand gestures. This training, coupled with the networks ability to generalize, allowed the network to correlate an input voltage profile with the gesture that generated it and correctly classify it.

Shape memory alloys are capable of delivering advantageous solutions to a wide range of engineering-based problems. Implementation of these solutions, however, is often complicated by the hysteretic, non-linear, thermo-mechanical behavior of the material. Although existing shape memory alloy constitutive models are largely accurate in describing this unique behavior, they require prior characterization of the material parameters. Consequently, before thorough modeling and simulation can occur for a shape memory alloy-based project, one must first go through the process of identifying several material parameters unique to shape memory alloys. Current characterization procedures necessitate extensive experimentation, data collection, and data processing. As a result, these methods simultaneously create a high barrier of entry for engineers new to active materials and impede the advanced study of shape memory alloy material parameter evolution. This paper develops a novel method in which computational intelligence methods are used to rapidly identify shape memory alloy material parameters. Specifically, an artificial neural network is trained to identify transformation temperatures and stress influence coefficients of given shape memory alloy specimens using strain-temperature coordinates as inputs. After generating training data through the use of a constitutive model, the resulting trained artificial neural network was used to identify parameters for a number of randomly generated theoretical shape memory alloys. Results presented in the paper show that the artificial neural network was able to rapidly identify both transformation temperatures and stress influence coefficients with satisfactory accuracy. The generation of training data was then repeated using Taguchi methods. Further results presented in the paper show that the artificial neural network trained with the Taguchi-based training data yielded improved characterization accuracy while using less training data.

In this paper the dependence of the torque characteristic of a magnetorheological clutch on several working parameters is analysed by means of a feedforward neural network. The clutch was envisaged to have the possibility to disengage the vacuum pump in diesel engine vehicles, in order to increase the overall vehicle efficiency. A large set of test was carried out following different protocols in order to obtain a detailed characterization of the clutch. Results showed how, due to the characteristics of MR fluids and to the complex mechanism of torque transmission, the torque characteristics (both the yield torque and the torque-slip behaviour) is function of the relative speed between the primary and secondary clutch groups and also function of the dissipated energy during the clutch slip and of the rest time between consecutive slippages. The data acquired during all tests were used to train a neural network with five input elements and one output element. The developed neural network was proved to satisfactorily reproduce the actual clutch properties and can be used in the future in an engine-vehicle simulator in order to model the torque characteristic of the clutch subjected to different operating cycles.

Delamination is a frequent and potentially serious damage that can occur in laminated polymer composites due to the poor inter-laminar fracture toughness of the matrix. Vibration based detection methods employ changes caused by loss of stiffness in dynamic parameters such as frequencies and mode shapes to detect and assess damage. Because it is a whole field method, and can be applied instantaneously and remotely, vibration monitoring using frequency measurements offers great potential for implementation in online structural health monitoring systems. However, one of the disadvantages of using frequency measurements is that while the presence of damage is easily identified through a shift in measured frequency, the determination of the location and the severity of the damage is not easy to accomplish. To determine the location and severity of damage from measured changes in frequency, it is necessary to solve the inverse problem, which requires the solution of a set of non-linear simultaneous equations. In this paper, we have compared the performance of three different inverse algorithms for delamination detection in the fibre-reinforced composite laminates: direct of solution using a graphical method, artificial neural network (ANN) and surrogate-based optimization. In particular, the graphical method which was earlier proposed for problems of two variables has been extended to solution of three variables, the interface, location along the beam length and size of delamination in laminated composite beams. The three inverse algorithms have been compared using numerical validation data generated from the theoretical model of delaminated beam with and without artificial errors. All three algorithms can predict the delamination parameters accurately using the validation data directly generated from theoretical model. However, if artificial errors are introduced in the numerical data to simulate uncertainties in measurement of frequencies, ANN does not fare as well as the the other two methods as it is more sensitive to the artificial discrepancies. Also, ANN requires the network to be retrained if the measured frequency modes do not match the input modes in the existing network. The graphical technique and the surrogate based optimization performed equally well in the validations. However, the graphical technique is only applicable to no more than three variables, while the surrogate-based optimization algorithm can be applied to inverse problems with several unknown parameters such as in the case of delaminations in composite plates.

This paper presents the development of an indirect intelligent sliding mode controller (IISMC) for shape memory alloy (SMA) actuators. The controller manipulates applied voltage, enabling temperature control in one or more SMA tendons, which are offset to produce bending in a flexible beam tip. Hysteresis compensation is achieved using a hysteretic recurrent neural network (HRNN), which maps the nonlinear, hysteretic relationships between SMA temperatures and bending angle. Incorporating this HRNN into a variable structure control architecture provides robustness to model uncertainties and parameter variations. Single input, single output and multivariable implementations of this control strategy are presented. Controller performance is evaluated using a flexible beam deflected by single and antagonistic SMA tendons. Experimental results demonstrate precise tracking of a variety of reference trajectories for both configurations, with superior performance compared to an optimized PI controller for each system. Additionally, the IISMC demonstrates robustness to parameter variations and disturbances.

With the aim to decrease the uncertainties of structural damage detection, two fusion models are presented in this paper. The first one is a weighted and selective fusion method for combing the multi-damage detection methods based on the integration of artificial neural network, Shannon entropy and Dempster-Shafer (D-S) theory. The second one is a D-S based approach for combing the damage detection results from multi-sensors data sets. Numerical study on the Binzhou Yellow River Highway Bridge and an experimental of a 20-bay rigid truss structure were carried out to validate the uncertainties decreasing ability of the proposed methods for structural damage detection. The results show that both of the methods proposed are useful to decrease the uncertainties of damage detection results.

In systems with hysteresis behavior like Shape Memory Alloy (SMA) actuators and Piezo actuators, an accurate modeling of hysteresis behavior either for performance evaluation and identification or controller design is essentially needed. One of the most interesting hysteresis none-linearity identification methods is Preisach model which the hysteresis is modeled by linear combination of hysteresis operators. In spite of good ability of the Preisach model to extract the main features of system with hysteresis behavior, due to its numerical nature, it is not convenient to use in real time control applications. In this paper a novel artificial neural network (ANN) approach based on the Preisach model is presented which provides accurate hysteresis none-linearity modeling. It is shown that the proposed approach can represent hysteresis behavior more accurately in compare with the classical Preisach model and can be used for many applications such as hysteresis non-linearity control, hysteresis identification and realization for performance evaluation in some physical systems such as magnetic and SMA materials. It is also greatly decrease the extremely large amount of calculation needed to numerically implement the Preisach hysteresis model. For evaluation of the proposed approach an experimental apparatus consists of one-dimensional flexible aluminum beam actuated with a SMA wire is used. It is shown that the proposed ANN based Preisach model can identify hysteresis none-linearity more accurately than the classical Preisach model besides to its reduction in the simulation and computation time.