Algoritmo para la reducción de tratamientos en diseños experimentales con mezclas de vértices extremos, basado en la recuperación de la forma regular de un simplex
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
La investigación que se desarrolla a continuación persigue como su objetivo el diseñar un algoritmo para la reducción de tratamientos en diseños experimentales con mezclas de vértices extremos generados por restricciones en la frontera superior, basado en la recuperación de la forma símplex, reduciendo a su vez el tamaño de la región de experimentación. El algoritmo es nombrado CONVEXSIM (CONvertir Vértices Extremos a SIMplex) y cuenta con dos variantes: I y S. Con estos se logra convertir en seis pasos a las restricciones originales en el límite superior en restricciones de frontera inferiores y superiores que decantan en una forma L – Símplex, reduciendo así el tamaño de la región de experimentación y la cantidad de tratamientos experimentales.
Mediante un estudio de simulación determinística, que tiene como insumo la variación de la cantidad de componentes (p), la cantidad de restricciones sobre la frontera superior y el tamaño de estas, el grado del polinomio (m) máximo que se puede estimar, así como la presencia o ausencia de puntos axiales y la cantidad total de réplicas; se concluye que el algoritmo conduce a la reducción en la cantidad de tratamientos, consiguiendo generar diseños que pueden ser eficaces y que la mayoría de las ocasiones son eficientes. El estudio del caso práctico permite demostrar que la principal desventaja de CONVEXSIM no es un obstáculo para la consecución de los objetivos de un programa de experimentación, obteniendo resultados razonables en función del contexto del experimento.
Finalmente, se subraya que el algoritmo diseñado no es la panacea, y que su principal desventaja recae en que al reducir la región de experimentación el óptimo global puede quedar fuera y no ser estudiado. Se recomienda que su uso esté sujeto al contexto del experimento y a la perspicacia de las personas experimentadoras para moverse dentro de la región de operabilidad mediante el uso de la experimentación secuencial.
The research developed below aims to design an algorithm for reducing treatments in experimental designs with mixtures of extreme vertices generated by upper boundary constraints, based on the recovery of the simplex form, while simultaneously reducing the size of the experimental region. The algorithm, named CONVEXSIM (CONVert EXtreme vertices to SIMplex), has two variants: I and S. Through a six-step process, these variants transform the original upper boundary constraints into both lower and upper boundary constraints, resulting in an L-Simplex form, which reduces the size of the experimental region and the number of experimental treatments. A deterministic simulation study was conducted using variables such as the number of components (p), the number and magnitude of upper boundary constraints, the degree (m) of the highest polynomial estimable, the presence or absence of axial points, and the total number of replicates. The findings demonstrate that the algorithm effectively reduces the number of treatments, generating designs that are not only practical but also, in most cases, efficient. A practical case study further demonstrates that the primary limitation of CONVEXSIM does not hinder achieving the objectives of an experimental program, yielding reasonable results within the context of the experiment. Finally, it is emphasized that the designed algorithm is not a panacea, as its main limitation lies in the possibility that reducing the experimental region might exclude the global optimum, leaving it unexplored. It is recommended that the algorithm's use be contextualized to the experiment and supported by the experimenters' expertise, leveraging sequential experimentation to navigate the operability region effectively.
The research developed below aims to design an algorithm for reducing treatments in experimental designs with mixtures of extreme vertices generated by upper boundary constraints, based on the recovery of the simplex form, while simultaneously reducing the size of the experimental region. The algorithm, named CONVEXSIM (CONVert EXtreme vertices to SIMplex), has two variants: I and S. Through a six-step process, these variants transform the original upper boundary constraints into both lower and upper boundary constraints, resulting in an L-Simplex form, which reduces the size of the experimental region and the number of experimental treatments. A deterministic simulation study was conducted using variables such as the number of components (p), the number and magnitude of upper boundary constraints, the degree (m) of the highest polynomial estimable, the presence or absence of axial points, and the total number of replicates. The findings demonstrate that the algorithm effectively reduces the number of treatments, generating designs that are not only practical but also, in most cases, efficient. A practical case study further demonstrates that the primary limitation of CONVEXSIM does not hinder achieving the objectives of an experimental program, yielding reasonable results within the context of the experiment. Finally, it is emphasized that the designed algorithm is not a panacea, as its main limitation lies in the possibility that reducing the experimental region might exclude the global optimum, leaving it unexplored. It is recommended that the algorithm's use be contextualized to the experiment and supported by the experimenters' expertise, leveraging sequential experimentation to navigate the operability region effectively.
Description
Keywords
Experimentos con mezclas, CONVEXSIM, Diseño de experimentos, Reducción de tratamientos experimentales, Vértices extremos, Estadística