Papers by Keyword: Bayesian Inference

Abstract: Accurately modeling the fatigue life and strength of additively manufactured (AM) components ensures their reliability and performance in critical applications. However, this task is hindered by the complexities of AM processes, including material defects, anisotropy, residual stress, and surface roughness. This review explores how integrating surrogate modeling, transfer learning, and Bayesian inference can address these challenges and elevate predictive capabilities to new levels of accuracy and robustness. Surrogate modeling offers computationally efficient approximations of the intricate relationships between AM process parameters and fatigue behavior, enabling rapid exploration and optimization of design spaces. Transfer learning facilitates the adaptation of knowledge across different machines and process conditions, improving predictions even in low-data scenarios. Bayesian inference adds a layer of reliability by incorporating uncertainty quantification and prior knowledge into the modeling process. Together, these advanced methodologies present a transformative opportunity to improve the quality, efficiency, and robustness of fatigue life predictions for AM components, setting the stage for their broader adoption in high-performance applications

Abstract: Advances in meta-modelling and increasing computational capacity of modern computerspermitted many researches to focus on parameter identification in probabilistic setting. Increasinglypopular Bayesian inference allows to estimate model parameters together with corresponding epistemicuncertainties from indirect experimental measurements. However in case of a heterogeneousmaterial model, the identification procedure has to be able to quantify the aleatory uncertainties capturingthe variability of the material properties. Parameter identification of a heterogeneous materialmodel can be formulated as a search for probabilistic description of its parameters providing the distributionof the model response corresponding to the distribution of the observed data, i.e. a stochasticinversion problem. By prescribing a specific type of probability distribution to the model parameterswith corresponding uncertain moments, the task changes to the identification of these so-calledhyperparameters of the distribution which can be inferred in the Bayesian way.

Abstract: Models are often used to make predictions far from the region where they were trained and validated. In this paper attempts are made to analyse the credibility that can be placed in such predictions. The proposed approach involves treating a model’s parameters as time-variant (even if it is believed that this is not the case), before utilising Bayesian tracking techniques to realise parameter estimates. An example is used to demonstrate that, relative to a Bayesian approach where the parameters are assumed to be time-invariant, treating the parameters as time-variant can reveal important flaws in the model and raise questions about its ability to make credible predictions.

Abstract: A novel growing-window recursive algorithm for stochastic system change detection is derived based on the Bayesian inference principle. Model based detectors can be formalized by two concepts in literature: (a) working in a sliding-window strategy because of time-dependent computational complexity, or (b) running in parallel, each one matched to a certain assumption on a change point. This motivates us to investigate a more refined approach which utilizes all relevant data to catch the next change point. The basic idea is to formulate a distance measure between two probabilities, one confirming the change occurrence and the other confirming no change in the system behavior. This study aims to solve the difficulty of sliding time arguments in the compared probabilities as new data are sequentially obtained. The outcome of this analysis is an algorithm that recognizes the time and magnitude of the change occurrence.

Abstract: The effective method for Bayesian unconditional Cramer-Rao bound on condition that the unknown state-vector of a dynamical system is constant has proposed. The recurrence formula for calculating the Fisher information matrix is deduced. Our formula doesn’t follow from the well-known recurrence relations for the general case, where the state-vector varies, and has some advantages compared to them. The effectiveness of the proposed recursive method has been illustrated by applying to Вearing-only tracking.

Abstract: Information theoretic metrics is popular theory to measure anonymity. However the difficulty in getting the probability distribution of subjects hampers its practical usage. In this paper we propose a Bayesian inference method to tackle this problem. Our method makes it possible to compare the anonymity of different anonymous systems. We use this method to analyze Threshold Mix and point out different system parameters which do and do not have influence on anonymity.

Abstract: Radio Frequency Identification (RFID) technologies are used in many applications for data collection. However, raw RFID readings are usually of low quality and may contain many anomalies. The solution should take advantage of the resulting data redundancy for data cleaning. In this paper we propose a Bayesian inference based approach for cleaning RFID raw data. Our approach takes full advantage of data redundancy. To capture the likelihood, we design a 3-state detection model and formally prove this model can maximize the system performance.

Abstract: Simply attributing incidents to human error is not adequate; human factors aspects should be investigated such that lessons are learnt and the true root causes are established in order to prevent recurrence. Whilst many petroleum and allied industry businesses have investigated and analyzed incidents – whether with major hazards or occupational injuries potential – human factors aspects are rarely addressed sufficiently. Therefore, this paper presents a hybrid methodology that combines a conventional Swiss Cheese model with Bayesian inference to predict the failure probability of human factors. An inherent safety concept associated with human factor is proposed and utilized as preventive measures to overcome the identified root causes. This approach is then applied to offshore safety assessment study. As a result, the failure probability of human factor can be monitored with time and the best preventive measure can be prioritized once human performance is degraded. It is proven that the approach has the ability to act as predictive tool that provides early warnings toward human deficiency. A preventive measure can then be taken to enhance the overall human performance and ultimately to reduce the likelihood of major incidents.

Abstract: The basic frequency of masonry specimens can be obtained by dynamic tests with ambient or artificial excitation. The elastic modulus of masonry structures, as well as the damage factors, can then be determined by training their finite element models and make the calculated frequencies agree with the measured ones. Using 530 groups of dynamic test data, the damage factors of four masonry specimens were identified. The Bayesian inferences of the highly diverse measured results were conducted through a Markov Chain Monte Carlo (MCMC) sampling method, and the location of the damage was identified. The methodology was applicable, and can be used in the damage identification for other materials or structures.

Abstract: This paper illustrates an application of Bayesian logic to monitoring data analysis and structural condition state inference. The case study is a 260 m long cable-stayed bridge spanning the Adige River 10 km north of the town of Trento, Italy. This is a statically indeterminate structure, having a composite steel-concrete deck, supported by 12 stay cables. Structural redundancy, possible relaxation losses and an as-built condition differing from design, suggest that long-term load redistribution between cables can be expected. To monitor load redistribution, the owner decided to install a monitoring system which combines built-on-site elasto-magnetic and fiber-optic sensors. In this note, we discuss a rational way to improve the accuracy of the load estimate from the EM sensors taking advantage of the FOS information. More specifically, we use a multi-sensor Bayesian data fusion approach which combines the information from the two sensing systems with the prior knowledge, including design information and the outcomes of laboratory calibration. Using the data acquired to date, we demonstrate that combining the two measurements allows a more accurate estimate of the cable load, to better than 50 kN.