Circular Image

P. Komninos

info

Please Note

9 records found

Doctoral thesis (2025) - P. Komninos, R.C. Alderliesten, D. Zarouchas
This dissertation introduces a comprehensive end-to-end framework related to Prognostics and Health Management (PHM) strategy, with the goal of utilizing raw sensor (placed on a structure's components) data to make maintenance decisions for extending the structure's lifecycle. The study addresses key challenges in PHM, such as dealing with high-dimensional, multi-modal data, extracting features, predicting Remaining Useful Life (RUL), managing uncertainty, modeling repair scenarios, and optimizing maintenance decisions. All these challenges are organized in phases. These phases are namely, i) data collection, ii) feature extraction, iii) Health Indicator (HI) construction, iv) prognostics, v) modeling of maintenance actions, and vi) Post-Prognosis Decision-Making (PPDM). Each phase of the PHM strategy is developed independently but integrated into a cohesive framework, ensuring modularity, transparency, and adaptability across diverse applications.

A major contribution to the framework is proving that neural networks, a preferable approach for handling complex data, can become interpretable, thus unveiling the black-box nature of such models. In this regard, the ISTRUST model, an interpretable Transformer-based architecture for predicting RUL directly from raw sequential image data of a structure under fatigue loads, is proposed. By leveraging attention mechanisms, the model captures critical spatiotemporal features of structural damage, offering insights into prediction accuracy and variability. The model's interpretability highlights the understanding of its predictions. Simultaneously, via this interpretation, it is shown that predicting RUL directly from high-dimensional raw data is challenging or even impossible, necessitating focusing on each phase of the PHM strategy separately instead of unifying the majority of the PHM strategy in one model.

Central to the framework is the development of the Deep Soft Monotonic Clustering (DSMC) model, designed to extract meaningful features and then construct HIs from multi-modal data. This model extracts monotonic features that are related to each component's degradation. Subsequently, it considers these features to perform monotonic clustering representing HIs and enables the integration of those HIs into prognostic models to predict RUL across diverse domains.

The dissertation further explores the impact of imperfect repairs on components, where repairs often leave the component in a state between fully restored and partially damaged. The health state of the component is measured via a stochastic recovery of the predicted RUL and a Bayesian inference-based model is employed to quantify this stochastic behavior. The imperfect repair (Bayesian) model can be also extended for multiple sequential repairs. This phase of the PHM strategy emphasizes the importance of understanding imperfect repair effectiveness to enhance maintenance strategies.

The final phase of the framework concerns PPDM. Given a set of maintenance actions, including replacements and imperfect repairs, and a set of operational conditions and constraints, PPDM is modeled as a Markov Decision Process and optimized via deep Reinforcement Learning. PPDM's ultimate target is to optimize the scheduling of maintenance actions proactively within a predefined horizon length.

Experimental validation is conducted on each of the proposed models. The ISTRUST model was validated on fatigue-loaded composite specimens, utilizing sequential raw image data taken by a camera. The DSMC model was tested using diverse datasets, including engineering and healthcare. The imperfect repair modeling was applied to tension-tension fatigue experiments on open-hole aluminum coupons, capturing stochastic recovery behavior after repairs. The same experiment was utilized for evaluating the PPDM framework.

Overall, a holistic end-to-end PHM framework lays the groundwork for advancing Condition-based Maintenance (CBM) strategies. By integrating advanced models for each phase of the PHM strategy, the research highlights practical opportunities for embedding PHM into CBM and encourages further innovation and refinement by other researchers in the field. Although the proposed PHM framework has been an initial attempt towards this direction, its capabilities can be further enhanced through improvements in data scalability, exploration of varied PPDM formulations, sensitivity analyses across PHM phases, and practical integration with CBM for broader system-level maintenance optimization. ...
Maintenance decisions often involve choosing between replacement and repair. The shortage of essential replacement parts has led to increased exploration of repair methodologies. However, repairs are often imperfect, leading to additional uncertainties in predicting the component's future condition. Existing approaches in the literature for modeling imperfect repairs struggle when repair dynamics are unknown requiring a large amount of data to be reliable. Furthermore, current methods are very task-specific, which limits the optimization of maintenance planning of varying components. This research addresses these challenges by conceptualizing imperfect repair effects as a stochastic increase in Remaining Useful Life (RUL). An existing deep learning model extracts prognostic-related features that can be utilized by any prognostic model to estimate RUL based on sensor data. Then, the proposed imperfect repair model predicts the RUL increase post-repair. This method offers three key benefits: (i) proactive post-repair assessment for improved maintenance, (ii) a data-driven repair model compatible with existing prognostic models, and (iii) flexibility in adapting to different repair techniques. Evaluation of the proposed model is conducted through tension-tension fatigue experiments on aerospace-grade aluminium specimens subject to imperfect repair. Results demonstrate the model's ability to accurately estimate the post-repair stochastic RUL increase. ...
Abstract: Remaining useful life predictions depend on the quality of health indicators (HIs) generated from condition monitoring sensors, evaluated by predefined prognostic metrics such as monotonicity, prognosability, and trendability. Constructing these HIs requires effective models capable of automatically selecting and fusing features from pertinent measurements, given the inherent noise in sensory data. While deep learning approaches have the potential to automatically extract features without the need for significant specialist knowledge, these features lack a clear (physical) interpretation. Furthermore, the evaluation metrics for HIs are nondifferentiable, limiting the application of supervised networks. This research aims to develop an intrinsically interpretable ANN, targeting qualified HIs with significantly lower complexity. A semi-supervised paradigm is employed, simulating labels inspired by the physics of progressive damage. This approach implicitly incorporates nondifferentiable criteria into the learning process. The architecture comprises additive and newly modified multiplicative layers that combine features to better represent the system’s characteristics. The developed multiplicative neurons are not restricted to pairwise actions, and they can also handle both division and multiplication. To extract a compact HI equation, making the model mathematically interpretable, the number of parameters is further reduced by discretizing the weights via a ternary set. This weight discretization simplifies the extracted equation while gently controlling the number of weights that should be overlooked. The developed methodology is specifically tailored to construct interpretable HIs for commercial turbofan engines, showcasing that the generated HIs are of high quality and interpretable. ...
Journal article (2025) - P. Komninos, D. Zarouchas
The digitalization era has introduced an abundance of data that can be harnessed to monitor and predict the health of structures. This paper presents a comprehensive framework for post-prognosis decision-making that utilizes deep reinforcement learning (DRL) to manage maintenance decisions on multi-component systems subject to imperfect repairs. The proposed framework integrates raw sensory data acquisition, feature extraction, prognostics, imperfect repair modeling, and decision-making. This integration considers all these tasks independent, promoting flexibility and paving the way for more advanced and adaptable maintenance solutions in real-world applications. The framework's effectiveness is demonstrated through a case study involving tension-tension fatigue experiments on open-hole aluminum coupons representing multiple dependent components, where the ability to make stochastic RUL estimations and schedule maintenance actions is evaluated. The results demonstrate that the framework can effectively extend the lifecycle of the system while accommodating uncertainties in maintenance actions. This work utilizes the Value of Information to choose the optimal times to acquire new data, resulting in computational efficiency and significant resource savings. Finally, it emphasizes the importance of decomposing uncertainty into epistemic and aleatoric to convert the total uncertainty into decision probabilities over the chosen actions, ensuring reliability and enhancing the interpretability of the DRL model. ...
Addressing and predicting degenerative phenomena in domains such as healthcare and engineering, two fundamental fields of vital importance for society, offers valuable insights into early warning steps and critical event forecasting, leading to far-reaching implications for safety and resource allocation. By harnessing the power of data-driven insights, prognostics becomes the principal component of predicting such phenomena. Developing clustering techniques as feature extractors acts as an intermediate step between the raw incoming data and prognostics and provides the opportunity to unveil hidden relationships within complex datasets. However, when limited, noisy, and multi-modal data are available in a label-free format, extensive preprocessing, and unreliable, complicated models are required for extracting meaningful features. This prohibits the development of adaptable methods in diverse domains that are in favor of robustness and interpretability. In this regard, this study introduces a novel unsupervised deep clustering model for feature extraction in degenerative phenomena. The model innovatively extracts prognostic-related features from raw data via clustering analysis, characterized by an increasing monotonic behavior representing system deterioration. This monotonicity is partial rather than complete, to incorporate the potential occurrence of oscillations in the degradation trajectory of the system or noise-related data, reflecting real-world scenarios. Its performance, robustness, generalizability, and interpretability are evaluated across diverse domains utilizing three datasets from healthcare and engineering featuring limited, noisy, high-dimensional, and multi-modal raw signals. Results show that the model extracts meaningful prognostic-related features in both domains and all datasets, without a significant alteration in its architecture and independently of the chosen prognostic algorithm. ...
Journal article (2024) - Dharun Vadugappatty Srinivasan, Morteza Moradi, Panagiotis Komninos, Dimitrios Zarouchas, Anastasios P. Vassilopoulos
In this research, a generalized machine learning (ML) framework is proposed to estimate the fatigue life of epoxy polymers and additively manufactured AlSi10Mg alloy materials, leveraging their failure surface void characteristics. An extreme gradient boosting algorithm-based ML framework encompassing Synthetic Minority Over-sampling TEchnique (SMOTE), categorical data encoding, and external loop cross-validation is developed to evaluate the fatigue life across materials. The influence of different training strategies based on materials, input features, encoding method, and data standardization on the model performance is explored. Additionally, the importance of anti-data-leakage and anti-overfitting measures over the ML model performance is addressed. The result shows that the data-leakage-free, external loop cross-validated model can estimate the fatigue life of selective epoxy polymers and metal alloys with an average R2 of 0.71 ± 0.06 using a mere 12 to 27 experimental data points per material category. Whereas the model trained with data-leakage and overfitting results in high R2 of 0.9. ...
In recent years, prognostics gained attention in various industries by optimizing maintenance, boosting operational efficiency, and preventing costly downtime. Central to prognostics is the Remaining Useful Life (RUL), representing the critical time before system failure. Deep learning advancements facilitate RUL forecasting by extracting features from diverse data formats such as time series, images, or sequences thereof, in one, two, or three dimensions, respectively. Yet, predicting RUL from image sequences often relies heavily on resource-intensive techniques like digital image correlation, complicating data acquisition. To address challenges with high-dimensional data and unreliable models, this study introduces ISTRUST, an innovative Transformer-based architecture. ISTRUST (Interpretable Spatiotemporal TRansformer for Understanding STructures) tackles the dual challenges posed by high-dimensional data and the black-box nature of existing models. Leveraging Transformers’ attention mechanism, ISTRUST breaks down the spatiotemporal domain, effectively realizing interpretable RUL predictions under uncertainty using only sparse raw image sequences as input. Evaluated on fatigue-loaded composite samples showcasing crack propagation, ISTRUST interprets the relation between cracks and RUL via the attention mechanism. The results substantiate its capacity to interpret and clarify instances in which predictions may exhibit variability in accuracy. Through the attention mechanism, a strong correlation between the model’s spatiotemporal focus and the RUL predictions is established, making it, to the best of our knowledge, the first model to provide interpretable stochastic RUL predictions directly from sequential images of this nature. ...
A hybrid methodology based on numerical and non-destructive experimental schemes, which is able to predict the structural level strength of composite laminates is proposed on the current work. The main objective is to predict the strength by substituting the up to failure experiments with non-destructive experiments where the investigated specimen is loaded up to 20% of its maximum load. A significant gap exists between the 20% and the 100% load which is proposed to be treated by high fidelity physics-based numerical models, deep learning techniques, and non-catastrophic experiments. Thus, a deep learning algorithm is developed, based on the convolutional neural networks and trained by probabilistic failure analysis datasets which result from the utilization of the stochastic finite element method. Also, the Monte Carlo dropout technique is embedded into the developed convolutional neural network to estimate the uncertainty induced by the investigated variations between the simulated and experimental data. The current paper provides a thorough description of the proposed methodology and a practical example which demonstrates the validity of the method. ...
Conference paper (2022) - M. Moradi, P. Komninos, R. Benedictus, D. Zarouchas
Recently, companies all over the world have been focusing on the improvement of autonomous health management systems in order to enhance performance and reduce downtime costs. To achieve this, the remaining useful life predictions have been given remarkable attention. These predictions depend on the proper designing process and the quality of health indicators (HI) generated from structural health monitoring sensors based on prior established multiple prognostic evaluation criteria. Constructing such HIs from noisy sensory data demands powerful models that enable the automatic selection and fusion of features taken from those relevant measurements. Deep learning models are promising to autonomously extract features in scenarios with a huge volume of data without requiring considerable domain expertise. Nonetheless, the features established by artificial neural networks are complicated to comprehend and cannot be regarded as physical system characteristics. In this regard, the goal of this paper is to extend a new model; an interpretable artificial neural network that enables the automatic selection and fusion of features to construct the most appropriate HIs with remarkably fewer parameters. This model consists of additive and multiplicative layers that provide a feature fusion that better reflects the system’s physical properties. Additionally, the weights are discretized in two ways: a) using a ternary form with values {-1, 0, 1}, and b) relaxing the aforementioned ternary form by rounding the weights at the first decimal point in the range of [-1, 1]. Both discretization techniques have the ability to softly control the number of parameters that should be ignored. This trick guarantees interpretability for the neural network by extracting simple yet powerful equations representing the constructed HIs. Finally, the model’s performance is evaluated and compared with other approaches using a practical case study. The results show that the proposed approach's designed HIs are both interpretable and of high quality according to the criteria of the HI's evaluation. ...