Modern manfacturing processes generate large datasets

Modern manufacturing processes generate large datasets, containing information gathered by the multiple sensors located in machines, buildings, products, etc. The management of this information is a challenge, since it implies the collection of diverse data sources and the compilation of different data formats, but also the understanding of the physical meaning of a huge variety of parameters and contexts. Beyond the practical challenge that this implies, it is now obvious to most of the operation managers using of Big Data that this is a great advantage for improving the firm’s operations, such as allowing the detection of patterns, developing new quality indicators, correcting existing deviations in the process, finding optimization opportunities, and so on.
Of course, data collection is nothing new in manufacturing processes, but the explosive growth of sensors over the last years has exponentially increased the number of data, making it impossible to manage all the outputs by means of traditional spreadsheets, which are unsuitable tools for handling high volumes of information, as they are slow, offer a low performance and require high maintenance.

Data Science tools (a combined approach of statistics, mathematics, programming and domain knowledge) offer not only the possibility to explore vast data sets but also to provide faster, more powerful tools and methods, offering a larger set of different approaches for handing our data, giving rise to a deeper understanding of process mechanisms and leading to important quality and productivity improvements.

All the above-mentioned reasons explain why the portion of the process engineer’s time devoted to researching the best tools for the management of the existing information is steadily increasing year by year, as is also the required background this kind of position requires, the findings derived from his/her research giving rise to new inputs for the process itself.

In recent years, several possible use cases of data analysis and machine learning models have evolved around the improvement of the manufacturing process, related to the different elements listed below:

Machine efficiency Increase:

  • Predictive Maintenance.
  • Higher process control.
  • Reduction of stop-overs.
  • Increase of productivity.

Product quality improvement:

  • Reduction of variability in key film characteristics.
  • Enhance traceability of metallized film.

To achieve the above-mentioned improvements levels in quality and efficiency in the most effective way, the use of external support is a common solution (for example from commercial solutions such as those from USU Software, Big ML, or even less specific software from Google or IBM).

In any case, three intermediate goals need to be successfully accomplished:

Developing an application for modeling of process parameters:

  • Analysis of main process parameters, using tools as Signal-to-Noise Ratio, Energy Analysis, Multivariate Analysis, scatter plots, histograms, process variation limits, etc.
  • Feature for browsing different batches, dates, etc.
  • Comparisons between different products possible, to execute a partial gap analysis of products sharing commonalities.
  • Implementation of the tool at plant / divisional level.

Identification of best quality indicators of the process:

  • FMEA using the previously mentioned tools for correlation between indicators and process/product quality.
  • Predictive analysis: identification of failure mode mechanisms.
  • Creation of documentation with knowledge acquired, in order to provide inputs for an online dynamic control system of the process.
  • Possibility to create a Process Batch Report, which could help reduce quality controls.
  • Online analysis of the key variables.
  • Comparison of the evolution of each key parameter during the particular run with the normal parameter window.
  • Early detection of deviations and creation of an alarm for the operator.
  • Comparison of the execution of the very same product in different machines.
  • A gap analysis, where the significant deviations in the key parameters will be identified.

Within a given organization, addressing all the above-mentioned points implies solving challenges other than the mere data science aspects. Most companies starting to use these new techniques will encounter many obstacles and limitations, related mainly to the difficulty of identifying the best process indicators (i.e. the existence of an accurate model for their process), or the complexity of the algorithms (giving rise to low performance of the applications.). Of course, there will also be some hardware and connectivity limitations (e.g. the existence of legacy systems) that need to be addressed for a successful deployment of any data analysis tool in an existing manufacturing process. A final and very important element to take into account, from a more organizational point of view, is that some resistance of film stakeholders to implementing the actions derived from the project are to be expected, as well as considering the time constraints to meet production deadlines. The well-known resistance to change of well-established organisations.

Nevertheless, even though the road may not run true nor be free of difficulties, the change of the manufacturing framework is an unstoppable reality, opening up new possibilities for business and carrier developments for our students.

By Ramón García-Rojo

Modern manfacturing processes generate large datasets was last modified: August 24th, 2018 by MIUC
Please follow and like us:

Be the first to comment

Leave a Reply

Your email address will not be published.



Enjoy this Website? Please spread the word :)

  • Facebook
  • Twitter
  • Google+
  • LinkedIn
  • Instagram
  • YouTube