实例探究 > Multivariate Statistical Analysis Finds the Bad Actors in Out-of-Spec Batches

Multivariate Statistical Analysis Finds the Bad Actors in Out-of-Spec Batches

产品

Aspen ProMV

技术栈

Machine Learning
Data Analysis

实施规模

Pilot projects

影响指标

Productivity Improvements
Cost Savings

技术

分析与建模 - 大数据分析

适用行业

化学品

适用功能

质量保证
离散制造

用例

质量预测分析
过程控制与优化

服务

数据科学服务

关于客户

The customer in this case study is a large producer of synthetic rubber. The company has been facing quality issues with its batch products, leading to significant revenue loss. The company often had to either reprocess the material or sell it at a lower price than expected due to these quality issues. The company was unable to determine the cause of these out-of-spec batches. The company has a long-standing relationship with AspenTech and uses a number of products from the aspenONE® Manufacturing and Supply Chain and Engineering suites.

登录后查看完整内容

挑战

A large producer of synthetic rubber had been having quality issues with its batch products. These quality issues were resulting in significant revenue loss, as the company often needed to either reprocess the material or sell it for a lower price than expected. The producer was unable to determine what was causing the batches to be out of spec. The company was investigating issues with a reactor process that brings together ingredients to manufacture synthetic rubber. There were multiple reactors that performed this process, but the Aspen ProMV project would focus on the production of one reactor.

登录后查看完整内容

解决方案

Aspen ProMV desktop batch model was developed to identify the bad actors in the off-spec batches. The customer provided five months of production data, representing 55 batches produced from this one reactor. Input variables included initial temperature, amount of catalyst and amount of other raw materials for each batch. Since this was a batch process, there was batch profile data (e.g., temperature, pressure, level, reactor agitator speeds, etc.) from the batch run. Quality variables measured at the end of each batch were also provided. There were three key quality variables that customer wanted to keep in control. The analysis was performed using Aspen ProMV desktop for batch. Aspen ProMV found several variables with very low variations and excluded them from the model.

登录后查看完整内容

运营影响

Aspen ProMV was able to highlight the few process variables (from a total of around 80) that correlate closely to quality.
Aspen ProMV showed how the company’s batch operating procedures were affecting batch quality.
Aspen ProMV proved its ability to show which variables correlate the most with batch quality.
Aspen ProMV identified which of these variables could be modified, and in which direction, to achieve better-quality batch products.

登录后查看完整内容