Papers by Keyword: Predictive Model

Abstract: This study presents a 250 kHz ultrasonic pulse velocity-based two-stage prediction model and a standardized field manual for diagnosing early frost damage in cementitious materials. The first stage predicts compressive strength from UPV and curing age, while the second stage estimates early frost damage depth using the predicted strength. Among several regression algorithms, ensemble models showed the highest predictive accuracy. Based on these results, a site-applicable standard operating procedure was developed, defining sampling rules, repeatability criteria, k-correction for indirect paths, and judgment protocols. The proposed model-to-manual framework enables fast, consistent, and reproducible on-site assessment of early frost damage during winter construction.

Abstract: During a heat treatment, a material undergoes microstructural changes that result in an alterationof its hardness. In a two-step heat treatment, the material is first adjusted to an initial hardness viaa specified cooling rate. Subsequently, the hardness is reduced through a tempering process, whileits ductility is increased. Depending on the tempering duration, tempering temperature, and initialhardness, different resulting hardness values are obtained. The resulting hardness after a chosen heattreatment has thus far been difficult to predict. This work employs symbolic regression to develop amodel that predicts the hardness evolution of 42CrMo4 steel as a function of cooling rate, temperingduration, and tempering temperature. By describing the model with few parameters, it has alsobeen demonstrated that cooling rates and tempering temperatures leading to a target hardness canbe determined. The overall model achieves a coefficient of determination of R2 = 98.50 % for knownexperimental data and a combined coefficient of determination of R2 = 93.13 % for previouslyunknown cooling rates (forward) and previously unattained resulting hardness values (inverse).Our work shows that the resulting hardness of 42CrMo4 can be predicted using a small numberof parameters. This work is anticipated to establish a foundation for further research endeavors.For instance, the approach using symbolic regression can be further adapted to identify physicallyinterpretable constants. Furthermore, the model description offers the possibility of coupling witha simulation model to accurately predict the hardness of a component.

Abstract: The paper develops a predictive model to determine the influence of the value of the dispersion component of the free surface energy (FSE) and the volume fraction of fillers on the dispersion component of the FSE of a polymer composite. Mathematical equations and graphical relationships illustrating these relationships are presented. The model is based on the assumption that in composites the FSE value is partially determined by interactions at the polymer-filler interface. Using the predictive model, it was established that the dispersion component of the free surface energy (FSE) can be a reflection of the properties of polymer composites. The reliability of the predictive assessment is shown on the example of an epoxy polymer composite with various mineral fillers.

Abstract: Generally, expansive soils undergoes significant volumetric deformation, which causes structural damages to existing infrastructures. Damages due to expansive activities are noticeable in pavements, buildings, earth dams, retaining walls etc. To estimate swelling stress, accurate assessment of soil absorption of water over time, with respect to soil volumetric change is required. However, the time frame requires for completion of swelling cycle is relatively long. With this in view, several attempts with great success have been made by researchers to predict swelling pressure of expansive soils using soil mechanics index properties. In this study, the interrelation between unsaturated soil mechanics property i.e. Matric suction () and geotechnical soil indexes were utilized to develop three predictive multi-regression equation for swelling stress. Series of Atterberg limit tests, matric suction tests, free swell index (FSI) tests and zero swelling tests (ZST) were performed to obtain the dependent and independent variables for the multi-regression analysis. Based on the experimental results, empirical relationships were developed to determine swelling stress as a function of matric suction, gravimetric moisture content (GMC), FSI, dry density and plasticity index using mathematical software package (NCSS11). The developed predictive multi-regression models were used to estimate the experimental swelling stress (. The scattered plot showed good agreement between the measured and predicted data, with coefficient of determination (R²) and mean square error (MSE) of 0.9443, 0.9793, 0.9310 and 0.0051%, 0.0021% and 0.0067% for models 1, 2 and 3 respectively.

Abstract: A new modeling method called multivariate adaptive regression spline (MARS) was firstly employed to predict the hot rolling flow stress and explain the relationship among flow stress and various parameters such as major chemical compositions, rolling temperature, rolling speed, compression ratio, thickness, roll radius, furthermore, analyze the importance of the predictor variables. The results showed that the error of training and testing was less than 2%, and rolling temperature, rolling speed, and strip thickness had much contribution to flow stress. Moreover, the impact of various factors on the flow stress can be validated by real production data, which proved the reliability of MARS model to predict the flow stress and guide the practical production.

Abstract: Polymer concrete is used for a wide range of precast structural applications and repair works for existing infrastructures. For these applications, one of the key mechanical propertiesthatneed to be consideredis the elastic modulus. In this study, the relationship between elastic modulus and compressive strength of polymer concrete made with three different types of resin (i.e., unsaturated polyester resin, acrylic resin, and epoxy resin) is comprehensively investigated using data sets available from previous studies in order to develop a prediction equation for elastic modulus that can be generally applied to polymer concrete. Results showed that the equation developed under this study can be reasonably adopted for the predictions of polymer concrete's elastic modulus as a function of compressive strength because the prediction equation has a high goodness of fit asrepresented by a R² value of 0.77

Abstract: Data-driven quality evaluation in the stamping process of car body parts is quite promising because dependencies in the process have not yet been sufficiently researched. However, the application of data mining methods for the process in stamping plants would require a large number of sample data sets. Today, acquiring these data represents a major challenge, because the necessary data are inadequately measured, recorded or stored. Thus, the preconditions for the sample data acquisition must first be created before being able to investigate any correlations. In addition, the process conditions change over time due to wear mechanisms. Therefore, the results do not remain valid and a constant data acquisition is required. In this publication, the current situation in stamping plants regarding the process robustness will be first discussed and the need for data-driven methods will be shown. Subsequently, the state of technology regarding the possibility of collecting the sample data sets for quality analysis in producing car body parts will be researched. At the end of this work, an overview will be provided concerning how this data collection was implemented at BMW as well as what kind of potential can be expected.

Abstract: To the question of the water level control of the underground drainage control system such as multi variables, tight coupling, nonlinear, and it was difficult to modulate precisely, the subsection control based on predictive model strategy of the pumps was proposed. The control rules were determined by analyzing the error which contain the predicted value for ensuring the stability of water level. Simulation shows that the algorithm is feasible, control system response time shortened, over regulation measurement greatly reduced, fluctuation time decreased and control system has strong adaptability and good stability.

Abstract: To predict the machining results of ultrasonic vibration grinding assisted electric discharge machining (UVGAEDM) in the condition of building predictive model with a few samples but fluctuant values, a predictive model based on SVM was proposed in this paper. Taking machining SiCp/Al as an example, the samples for modeling were obtained through orthogonal test, and then the predictive model was established utilizing MATLAB. Finally, the model was optimized to further improve the prediction accuracy about the processing indicatorssurface roughness and processing velocity. It shows that the predictive results are in good accord with the test results, with the maximum relative error being less than 12%, meaning the predictive model is reliable and effective.

Abstract: The thermal management is crucial to the safety and lifespan of Solid Oxide Fuel Cell (SOFC) generation system. For the model-predictive control design, a model of SOFC thermal management system is proposed on the least squares support vector machine (LS-SVM). The model is composed of some thermal modules including SOFC stack, combustor, heat-exchanger and thermal equilibrium apparatus. It predicts the temperature distribution in SOFC generation system by computing the electrochemical reaction in the stack, the gas flow and the heat exchange through the modules. Checked by the experimental data, the model can be used for system temperature fast prediction with high precision and strong generalization ability, which meets the requirement of the research on the online predictive control design of SOFC generation system.