UptimAI uses a proprietary AI algorithm for calculations and then provides clear outputs needed for detailed analysis of engineering tasks. You can read more about the features in this section:
The core of our solution is designed to save the time of engineers and help them to find innovative solutions. Our advanced AI algorithm uses the results of simulations to create a surrogate model of the problem.
The so-called metamodel is used for further analyses of the problem instead of simulations. Therefore, even complex problems can be solved with better accuracy and in just a fraction of time in comparison to standard approaches or Design Of Experiment methods.
The predictive scheme searches effectively through the design space and identifies sections most sensitive to changes in input parameters. Thus, the algorithm can focus on the most interesting subjects of the problem and there is no need for an excessive amount of simulation runs required to build the metamodel. The algorithm is not prone to the Curse of Dimensionality phenomenon and makes a project with dozens and more input parameters feasible to be solved.
The adaptive scheme is here to sample the design space with maximal effectivity. It adapts to the problem being solved and chooses sampling and interpolation methods on the fly to reach the best fit. Bespoke ISI (Independent Surrogate Interpolation) method can be used to interpolate complex non-linear and discontinuous problems to prevent the metamodel from divergence.
Understanding the ability of input parameters to affect the designed product is the first step to its improvement. The sensitivity analysis describes the importance of changes in geometry, material properties or manufacturing tolerances to the product performance. Based on the outputs of the sensitivity analysis, it is possible to focus on the most influential inputs to achieve the optimal design.
UptimAI works with the standard sensitivity analysis based on examination of the variance of results. Moreover, it also evaluates the sensitivity of the mean value to changes in inputs. Understanding the nature of the influence of each input helps to choose the right way leading to stable improvement in specific aspects of the product’s performance.
The input parameter can affect the output either on its own or due to interactions with other inputs. UptimAI sensitivity analysis can investigate these two categories separately by evaluation of sensitivity to increment functions. The increment represents an independent contribution of a given variable or interaction to the problem without being affected by the influence of other variables. Sensitivity analysis of increment functions evaluates both mean and variance sensitivity.
Standard and incremental plot
The dependence of outputs on changes in input parameters can be easily understood from standard and incremental function plots. UptimAI offers 2D, 3D, and 4D plots, where the 4D plot presents and explains interactions of 3 inputs. Examination of increment function shapes provides the key to the solved problem, allows to walk through the design space in detail, and eliminates weak points of the solution.
The standard plot of UptimAI shows the product behavior in the N-D space. It also includes two types of increment function plots. The increment function represents an independent contribution of a given variable or interaction to the problem without being affected by the influence of other variables. Increments simplify orientation in the nD design space making the problem more understandable.
The first type of increment plot shows the increment of the function value based on changes in an input parameter or combination of input parameters in case of interacting inputs. This type of plot can be easily used to search for design points where the simulation software failed to converge or gave an obviously wrong result.
The second type of increment function, the “Total Increment Plot”, shows the combined effect of selected inputs. Increment functions are additive, thus, the overall influence of interacting inputs can be shown as a sum of all their increments.
Using histogram plots allows displaying the nature of the influence of input parameters. UptimAI clearly shows changes in the probability distribution of results induced by each increment function, variable, or combination of variables to give a detailed view of their statistical impact. This additional insight helps to explain the product’s behavior under real conditions.
UptimAI composes the probability density function (probability distribution of results) from partial distributions of increment functions. These partial distributions are additive and can be plotted separately or grouped in arbitrary sets. It allows a deeper understanding of the statistical influence of each input parameter. As an example, some inputs may induce changes in the shape of the density function while the others shift the whole distribution towards a certain value.
The UptimAI Influencer is another feature showing directly the change in the final probability density function when uncertainties in selected increments or inputs are omitted from results. Thus, extremely adverse results and their causes on the side of inputs can be discovered and handled in the early stages of the design.
UptimAI provides a fundamentally new approach to optimization. Instead of a single optimized point, it recommends ranges of input parameters always leading to an increase in the performance of the product under all operating conditions. UptimAI optimization does not require any mathematical expertise, even for multidisciplinary optimizations due to no need for the definition of complex boundary conditions.
The process of preliminary optimization is based on statistics, hence, it manipulates ranges of input parameters to improve the solution. As the output, it provides a probability distribution where all values show enhanced product performance. It increases the overall probability of improvement and decreases the number of possible product failures.
UptimAI detects values in the range of each input variable that are more likely to achieve better results. Concurrently, it identifies the range of input values which needs to be avoided in order to prevent the optimized characteristics from the deterioration. It is also possible to generate sets of input ranges based on different criteria and compare them against each other to obtain the multiobjective optimum.
Based on simulations’ results, UptimAI builds a mathematical model describing relations between input parameters and outputs defined for the problem. This metamodel is created by the UptimAI algorithm to reach the best fit while minimizing the number of simulation runs. For further analyses of the problem, the metamodel is used instead of additional simulations to increase the effectiveness of the design process.
UptimAI, for instance, automatically creates the surrogate model from complex numerical simulations, each taking hours to finish. Then, it takes only a fraction of a second to obtain results from the metamodel. Using metamodels instead of lengthy numerical computations significantly reduces the simulation process and its costs.
The metamodel can be easily coupled to any type of software or script. It is the user’s choice which type of analysis will be performed on the model ensuring great variability of its use. The metamodel can be shared across the organization and used to exploit all the possible know-how from it.