Predictive maintenance, as one of the core components of Industry 4.0, takes a proactive approach to maintain machines and systems in good order to keep downtime to a minimum and the airline maintenance industry is not an exception to this. To achieve this goal, practices in Structural Health Monitoring (SHM) complement the existing Non-Destructive-Testing (NDT) have been established in the last decades. Recently, the increasing computational capability such as utilization of a graphical processing unit (GPU) in combination with advanced machine learning techniques such as deep learning has been one of the main drivers in the advancement of predictive analytics in condition monitoring. In our previous work, we proposed a novel approach using deep learning for guided wave based structural health called DeepSHM. As a study case, we treated an ultrasonic signal from guided Lamb wave SHM with a convolutional neural network (CNN). In that work, we only considered a single central frequency excitation. This led to a single governing wavelength which is normally good for the detection of a single damage size. In classical signal processing, applying a broader excitation frequency poses an analysis and interpretation nightmare because it contains more complex information and thus is difficult to understand. This problem can be overcome with deep learning; however, it creates another problem: while deep learning typically results in a more accurate result prediction, it is specifically made for solving only certain types of tasks. While many papers have already introduced deep learning for diagnostics, many of these works are only proposing novel predictive techniques, however the mathematical formalization is lacking, and we are not informed about why we should treat acoustic signal with deep learning. So, the basis of ‘explainable AI’ for SHM and NDT is currently lacking. For this reason, in this paper, we would like to extend our previous work into a more generalized. Rather than focusing on a novel technique, we propose a plausible theoretical perspective inspired from neuroscience for signal representation of deep learning framework to model machine perception in structural health monitoring (SHM), especially because SHM typically involves multiple sensory input from different sensing locations. To do this, we created a set of artificial data from a finite element model (FEM) and represented DeepSHM in two different ways: 1). Perpetual representation of observation and 2). Hierarchical structure of entities that is decomposable in a smaller sub-entity. Consequently, we assume two plausible models for DeepSHM: 1). Either it behaves as a single deciding actor since the observation is regarded as perpetual, and 2). Or it acts as a multiple actor with independent outputs since multiple sensors can form different output probabilities. These artificial data were split into several different input representations, classified into several damage scenarios and then trained with commonly used deep learning training parameters. We compare the performance metrics of each perception model to describe the training behavior of both representations.@en

One of the classical solutions to maintain the aircraft structural integrity is to rely on the analysis of non-destructive testing (NDT) inspector with various inspection methods. However, it is relatively expensive in matter of time and costs to train human resources until the certification is reached. Further, in majority of the cases of aircraft scheduled and unscheduled maintenance, most of the detected damages are far below the damage tolerance limit and therefore are considered as a costly false positive because such inspections generally require additional downtime. Structural Health Monitoring (SHM) tries to reduce the wasteful resources in the maintenance, repair, and overhaul (MRO) industry by signaling such false positives during the maintenance process by becoming an integral part of the structure itself. On the other hand, there has been an increase in using the artificial intelligence (AI) methodologies such as computational heuristics and machine learning in many areas of human civilization which includes voice and face recognition, languages translation, and automated driving. There has been a lot of interest on implementing AI to assist SHM in maintaining airworthiness while driving the cost down. Nevertheless, the maintenance of airworthiness (such as but not limited to, EASA Part 145/M and FAA CFR Part 21) is a heavily regulated area and are not easily changed. The current state of the art was captured in the literature review. This includes recent developments of guided wave based SHM and the parameter optimization as well as recent trends and advances in artificial intelligence such as machine and deep learning. The findings from the state of the art were used as the basis to determine the research problem and to propose the solution. The first part of the proposed solution consisted of a short review the damage growth assumption within the damage tolerance framework and the used methodology to generate and capture Lamb wave signal within Finite Element (FE) environment. This methodology is a deterministic solution that can be partially used for solving continuous optimization in deterministic sensor placement problem. It was further expanded to include a semi-stochastic approach to address nonpredictable damage location that includes some metaheuristics search such as genetic algorithm and swarm intelligence. The ultimate first part of solution was a compromise between the deterministic and semi-stochastic actuator-sensor topology. The second part of the proposed solution was the investigation on whether deep learning can be used to treat the Lamb wave signal given the configuration obtained from the first part of the proposed solution. To do so, an assumption based on converging probability measures and generalization bound in deep learning must be taken. Then, the approach is to represent the entity of the captured Lamb wave signal in time-frequency domain either as randomly sampled spectrogram or layers of joined spectrograms. After the training, the hypothesis was validated with A/B Testing. Then, the research was expanded to understand the scalability level of deep learning for SHM for given data size, model parameters, and restriction on physical memory. In this sense, the signal representations were trained sequentially with an example of in hybrid convolutional recurrent network. The investigation was focused on stability behavior of convoluted-recurrent modelling for variable spectrogram length and the experimental validation of the model for classification of the Lamb wave spectrogram signals. @en

In our previous work, we demonstrated how to use inductive bias to infuse a convolutional neural network (CNN) with domain knowledge from fatigue analysis for aircraft visual NDE. We extend this concept to SHM and therefore in this paper, we present a novel framework called DeepSHM which involves data augmentation of captured sensor signals and formalizes a generic method for end-to-end deep learning for SHM. The study case is limited to ultrasonic guided waves SHM. The sensor signal response from a Finite-Element-Model (FEM) is pre-processed through wavelet transform to obtain the wavelet coefficient matrix (WCM), which is then fed into the CNN to be trained to obtain the neural weights. In this paper, we present the results of our investigation on CNN complexities that is needed to model the sensor signals based on simulation and experimental testing within the framework of DeepSHM concept.@en

The near-term artificial intelligence, commonly referred as ‘weak AI’ in the last couple years was achieved thanks to the advances in machine learning (ML), particularly deep learning, which has currently the best in-class performance outperforming other machine learning algorithms. In the deep learning framework, many natural tasks such as object, image, and speech recognition that were impossible to be performed by classical ML algorithms in the previous decades can now be be done by typical home personal computer. Deep learning requires large amount of data that has to be rapidly collected (also known as ‘big data’) in order to create robust model parameters that are able to predict future occurrences of certain event. In some domains, a large dataset such as CIFAR-10, MNIST, or Kaggle exist already. However, in many other domains such as aircraft visual inspection, such a large dataset is not easily available and this clearly restricts deep learning to perform well to recognize material damage in aircraft structures. As many computer science researchers believe, we also think that in order to achieve a performance similar to human-level intelligence, AI could and should not start from scratch. Introducing an inductive bias into deep learning might be one solution to achieve that humanlevel intelligence. In this paper, we give an example how to incorporate aerospace domain knowledge into the development of deep learning algorithms. We performed a relatively simple procedure: we conducted fatigue testing of an aluminum plate that is typically used in aircraft fuselage and build a deep convolutional neural network that classifies crack length according to crack propagation curve obtained from fatigue test. The results of this network are then compared to the results of the same network that was not injected by domain knowledge @en

The concept of structural health monitoring has been introduced to ensure structural integrity during the design lifetime of a structure. The main objectives of structural health monitoring are to detect, locate, quantify, and predict any damage that occurs during this lifetime of the structure so that effective and efficient maintenance and repair procedures can be performed. The location of structural damage events can be discretized as deterministic and probabilistic. A deterministic location specifies that the damage occurs in high-stress regions or other regions that can be predicted by the structural design, such as the most probable location for a fatigue crack. A probabilistic damage event is one where the location of the damage is independent of structural design parameters, such as hail impact, bird strike, and impact from ground vehicles. A structural health monitoring system should be able to handle both these damage occurrences. In our previous work, we optimized the transducer placement in Lamb wave–based structural health monitoring for the detection of a fatigue crack that emerges from a rivet hole. In this article, we demonstrate a combination of that method with a different sensor placement optimization method to add the capability to detect probabilistic damage location. First, we considered the ultrasonic wave attenuation in the structure and based on this attenuation, we created a fitness function. Since this fitness function is difficult to solve due to its combinatorial nature, we compared three common metaheuristic stochastic strategies: global random search, greedy algorithm, and genetic algorithm, for solving this problem. The results of this analysis were then integrated with the previously described deterministic approach, making a global structural health monitoring sensor placement strategy that balances the need to detect both pre-determined and random damage location occurrences. The analytical result of the study presented is validated by experiment.@en

In this paper, we investigated transducer placement strategies for detecting cracks in primary aircraft structures using ultrasonic Structural Health Monitoring (SHM). The approach developed is for an expected damage location based on fracture mechanics, for example fatigue crack growth in a high stress location. To assess the performance of the developed approach, finite-element (FE) modelling of a damage-tolerant aluminum fuselage has been performed by introducing an artificial crack at a rivet hole into the structural FE model and assessing its influence on the Lamb wave propagation, compared to a baseline measurement simulation. The efficient practical sensor position was determined from the largest change in area that is covered by reflected and missing wave scatter using an additive color model. Blob detection algorithms were employed to determine the boundaries of this area and to calculate the blob centroid. To demonstrate that the technique can be generalized, the results from different crack lengths and from tilted crack are also presented. @en

In this paper, we review two transducer placement options to locate and quantify damage in primary aircraft structures using ultrasonic Structural Health Monitoring (SHM). The first placement approach concerns a known expected damage location, for example a fatigue crack growth from rivet hole. The location of such a damage can already be predicted by fracture mechanics and therefore the focus of this SHM system design is to determine the damage size. For this approach, we have developed our previous work in finite-element (FE) modelling of a damage tolerant aluminum fuselage by introducing an artificial crack into the structural FE model and assessed its influence on the Lamb wave propagation. Image processing was performed by subtracting the wave propagation image of the damaged from the undamaged structure. A second category of damage occurs at locations that cannot be predicted by fracture mechanics, such as impact damage from hail. This type of damage requires the SHM system to both locate and assess the size of the damage and this is heavily influenced by the positioning of the transducers. Optimal sensor placement (OSP) techniques tend to rely on assessment using the probability of detection (POD) parameter. In this work, we propose an alternative placement method which maximizes the detectability of the transducer coverage area based on the pulse-echo technique without relying on the POD parameter, by determining the fitness function based on sensor coverage area for single and multiple sensors and random damage locations. Results from both these approaches are compared in this paper, with a perspective towards the overall design of SHM systems.@en