This article is focused on areas where analytics are being used close to the production environment. Manufacturing companies obviously have other areas like marketing and customer relations etc. where they can benefit from insights generated from data as well.
Examples for analytics close to production are:
- In research and development, to design experiments efficiently and to draw conclusions from results.
- Monitoring of production processes to be informed about deviations from expected or stable conditions.
- In quality control to get the most precise insights about the quality of given products and batches.
In all these areas methods from “traditional” statistical analysis are being applied. This is a proven practice for many decades, and it is mandatory for any data-driven company. Those traditional methods are excellent tools, but they have certain limitations regarding their ability to respond to highly complex relationships in the data.
The methods from Machine Learning and AI allow to 1. deepen and 2. expand the spectrum of relationships understood from the data.
1. Deepen
Multidimensional relationships can be modelled in complex non-linear scenarios.
Machine Learning and AI can identify and process relationships that have previously not been understood, and where the complexity was beyond the capabilities of traditional methods.
For example:
- There are conditional relationships and other dependencies in the data, that cannot be automatically described with a simplistic model or where the manual modelling process would be too labor intensive.
There could for example be certain machine parameters that depending on machine condition have varying influence (sometimes positive, sometimes negative)
- The sum or the interaction of many weak signals are important when interpreted together.
This could be the aggregation of temperatures collected over multiple sensors, which might be more relevant than the individual readings.
This makes it possible to get more insights from existing data and to improve established analytical processes.
There are two ways of integrating these:
- By integrating the new ML or AI models in production and directly using their results.
- By taking the insights uncovered by the ML or AI models and integrating them in the traditional analytical process (for example: the sum of temperatures identified by the AI model could be fed to the models via feature engineering).
2. Expand
Challenges can be approached with ways of overcoming them. For example:
Image recognition could be used to detect defects.
Machine Learning and AI let us tap into data sources previously considered unusable and use them analytically. Data sources where the quality of the data was deemed too low could be used with more robust methods to be analyzed in combination with existing data. When for example raw sensor data is to be analyzed (for example with clean data from QA) it can often be seen that raw data does not meet the strict requirements of traditional statistical methods (regarding for example: outliers, distribution, multi-collinearity etc.) In those cases, ML models can be useful due to their less strict requirements.
Additionally, it is now possible to use information sources that were never even considered for analytical processes.
For example:
- Textual error protocols (potentially hand-written) can be made available via OCR (Optical Character Recognition) and Text Mining (NLP, LLMs, etc.,).
- Images from industrial cameras can be used via Image Recognition (CNN, Deep Learning, etc.)
Especially the domain of image recognition offers the potential for solutions that have been previously much too complex to solve (potentially only by using a multitude of sensors or only via destructive testing).
We as StatSoft are convinced that traditional analytical methods have their place and will not be replaced any time soon. But we believe that deepening and expanding via Machine Learning and AI can be a logical and valuable next step for our customers.
Please get in touch, so we discuss what ML and AI could mean for you and how you could benefit!
