As an expert in industrial quality, Ellistat is enhancing its module Data Analysis new artificial intelligence capabilities. In addition to machine learning and classification, the integration of neural networks makes it possible to model more complex industrial behaviors. With this software evolution, Ellistat has opted for an applied, gradual, and controlled approach to AI.
Understanding the causes of discrepancies and taking action on the right metrics is no longer just an intention—it is now a concrete approach.
Useful industrial AI, not just an abstract promise
For Ellistat, artificial intelligence is only valuable if it actually helps manufacturers
to produce better results. And to do that, it must first be accessible to everyone. Quality depends on
based on a simple principle: mastering one's processes, that is, understanding the parameters
factors that affect production: temperature, tool, material, machine, operator,
supplier, configuration, environment. But identifying these factors usually requires
statistical skills that not everyone has, and that few workshops
have time to mobilize.
When a deviation, discrepancy, or variation occurs, identifying the cause often involves
about the investigation: we proceed based on intuition, testing one parameter after another, and time
Time spent searching results in waste, delays, and adjustments that are more like
experience rather than evidence.
That's where Data Analysis comes in: the software handles the complex part, the
calculations, models, statistical methods, and presents it through an interface
Simple. You don't need to be a data science expert to analyze your data and identify the
influential variables, understanding the causes of variations, and building models
that can be used in the field. This gives everyone access to analytical capabilities that were previously
intended for specialists only.
This accessibility also defines the software's target audience. Data Analysis is not limited to
experts in statistics or artificial intelligence: it is primarily intended for those who
are familiar with the process: process engineers, production staff, and quality control staff. They are the ones who
have the on-the-ground knowledge, and it is for them that the tool must provide the analysis
accessible.
The new AI features follow this logic exactly: they add
power without sacrificing ease of use. For a non-expert user,
The approach remains guided, step by step. For an expert profile, the software offers more options
in-depth, with advanced settings for model structure and functions
activation: the user can configure a neural network, choose a training mode,
adjust how the model learns from the data or add neurons
depending on the desired level of analysis. The same interface thus adapts to the level of
each one, from rapid diagnosis to detailed modeling.
«Industrial AI shouldn't be just another black box on the factory floor. It should help the
teams to understand what's really happening in their processes. With DATA
Analysis: We want to provide manufacturers with more powerful models, but
useful to those who know the field. That's how AI becomes useful:
»when it informs production decisions," explains Davy Pillet, CEO of Ellistat.
From Classical Statistics to AI Models
The strength of Data Analysis lies in the wide range of methods it brings together in a single tool. At the
Basically, classical statistics address the most common questions in the workshop: comparing two variables, testing whether a change has a real effect, measuring a correlation, and constructing a regression or a design of experiments. These methods are simple to implement, remain indispensable, and form the foundation of a rigorous quality approach.
The new version significantly expands this range. Data Analysis includes
now includes machine learning models, neural networks, and
unsupervised classification. Whereas classical statistics describe simple relationships
Using a few variables, these models address phenomena in which many
The parameters interact: they determine the function that links the production conditions to the
the results obtained, and reveal patterns or behaviors that the eye
Humans don't distinguish between them.
This gives users a comprehensive range of options: they can choose the level of analysis that best suits
his question: from the two-variable test to the predictive model, without ever straying from the same
environment. The goal remains the same at every level: to reduce analysis time,
improve the reliability of diagnoses and focus efforts on the right levers.
At the Heart of the Ellistat Quality Suite
DATA Analysis is part of the Ellistat Quality Suite, a fully web-based software suite dedicated to
industrial-grade quality. The module truly shines when it leverages data from
other software components: with SPC or APC to track the statistical trends of their
processes and monitor production in real time and IQC to compare suppliers,
analyze the quality of incoming shipments or track changes in supplier performance
back then.
This complementarity ensures consistency across the entire Quality Suite: collecting
collect data, monitor it, analyze it, and then make decisions based on facts.
Progressive and Controlled Industrial AI
With this evolution in data analysis, Ellistat advocates a pragmatic approach to
Artificial intelligence. There can be no promise of autonomy without reliable data. No rhetoric disconnected from the realities of production. AI must be based on reliable data and processes
clear and measurable objectives.
Ellistat has chosen an AI solution that helps manufacturers better analyze, better understand, and
better manage their processes.
FAQ
Do you have to be a data scientist to use DATA Analysis?
No. DATA Analysis is primarily intended for process, production, and quality technicians who are familiar with the process. The software handles the calculations and models and guides the user step by step. Experts can, if they wish, access advanced settings: neural network structure, activation functions, and training mode.
What is the difference between classical statistics and machine learning?
Traditional statistics answer specific questions: Does a change have a real effect? Are two variables correlated? It remains the best approach when studying just a few parameters. Machine learning and neural networks take over when numerous parameters interact and the relationships are no longer easy to describe. In DATA Analysis, both approaches coexist within the same environment: you choose the level of analysis best suited to the question at hand.
What data is needed to train a model?
We need outcome data (a dimension, a scrap rate, a measured characteristic) and the production parameters that may influence them: machine, tool, material batch, operator, settings, and environmental conditions. The required data volume depends on the number of parameters studied and the complexity of the phenomenon. Quality is just as important as quantity: reliable measurements, taken with calibrated instruments, are essential for a usable model.
What is the purpose of unsupervised classification in production?
It automatically groups items that behave similarly, without requiring the groups to be defined in advance. In practice, this allows you, for example, to identify families of lots, machines, or suppliers with similar profiles, or to pinpoint atypical behavior that had not been identified previously.
Does DATA Analysis work with the other Ellistat modules?
Yes. DATA Analysis is most valuable when it leverages the data collected by the other modules in the Quality Suite: SPC for process history, IQC for supplier performance and the quality of incoming batches, and APC for machine measurements and adjustments. The data is already structured, which eliminates the need for a time-consuming preparation step.
Can an AI model replace the expertise of teams?
No. A model highlights the parameters that have the greatest impact on a result, but a correlation does not always imply a cause. Knowledge of the process remains essential for interpreting the results and deciding on next steps. When doubt remains, a design of experiments can confirm the actual effect of a parameter before modifying the process.
