AI as Analytical Consultant: The StatSoft GPT Connector in Manufacturing Practice

A new role for AI in manufacturing

In indus­tri­al analytics, arti­fi­ci­al intel­li­gence can assist ear­ly in the ana­ly­ti­cal pro­cess, at the point whe­re ana­lysts face the ques­ti­on: Which method is actual­ly the right one here?

The Stat­Soft GPT Con­nec­tor makes exact­ly this pos­si­ble. As a node in Sta­tis­ti­ca Workspaces, it con­nects tabu­lar pro­cess data and a pro­blem descrip­ti­on with an LLM, retur­ning a con­cre­te ana­ly­ti­cal plan tail­o­red to the available data and the spe­ci­fic qua­li­ty issue at hand.

A practical example

A qua­li­ty engi­neer in a che­mi­cal pro­duc­tion faci­li­ty is working with a data­set of 40 colum­ns cove­ring raw mate­ri­al pro­per­ties, pro­cess para­me­ters across four pro­duc­tion stages, and three final qua­li­ty cha­rac­te­ristics. A spe­ci­fi­ca­ti­on excee­dance has occur­red: the colour index of a spe­cial­ty poly­mer has repea­ted­ly fal­len out­side the per­mit­ted limits.

She opens a Sta­tis­ti­ca Workspace, con­nects the data­set to the GPT Con­nec­tor, and descri­bes the pro­blem:

“This data­set covers 18 months of batch pro­duc­tion data. The colum­ns include raw mate­ri­al mea­su­re­ments (vis­co­si­ty, mois­tu­re), reac­tor para­me­ters (tem­pe­ra­tu­re, pres­su­re, resi­dence time, agi­ta­ti­on speed), and final pro­duct pro­per­ties. What ana­ly­ti­cal approach would you recom­mend to iden­ti­fy the cau­ses of the colour index increase, and how could this be imple­men­ted in Sta­tis­ti­ca?”

The LLM recom­mends start­ing with a cor­re­la­ti­on and scat­ter plot matrix to screen all pro­cess varia­bles against the colour index. As a next step it sug­gests a Ran­dom Forest varia­ble importance ana­ly­sis to rank the most influ­en­ti­al pre­dic­tors. It notes that time-depen­dent effects should be inves­ti­ga­ted using con­trol charts, and high­lights batch-to-batch varia­bi­li­ty in the raw mate­ri­al mois­tu­re as par­ti­cu­lar­ly worth exami­ning. For imple­men­ta­ti­on in Sta­tis­ti­ca it points to the Data Mining modu­le and the Gene­ral Regres­si­on Models node.

Within minu­tes, the engi­neer has a struc­tu­red ana­ly­ti­cal plan, one that pre­vious­ly would have requi­red eit­her deep per­so­nal exper­ti­se or an exter­nal con­sul­ting enga­ge­ment.

Data privacy and flexibility

One prac­ti­cal advan­ta­ge of this approach is that it pri­ma­ri­ly trans­mits column names and a pro­blem descrip­ti­on rather than actu­al mea­su­re­ment values, allo­wing the flow of infor­ma­ti­on to the cloud to be careful­ly mana­ged. For orga­ni­sa­ti­ons with strict data gover­nan­ce requi­re­ments, the GPT Con­nec­tor also sup­ports local models via Oll­ama or GPT4All, kee­ping the enti­re pro­cess within the orga­ni­sa­ti­on’s own infra­struc­tu­re.

Looking ahead

As the reaso­ning capa­bi­li­ties of modern LLMs con­ti­nue to impro­ve, the pre­cis­i­on of the­se recom­men­da­ti­ons will increase fur­ther. Stat­Soft is con­ti­nuous­ly deve­lo­ping the Con­nec­tor and the broa­der eco­sys­tem of AI-enab­led nodes in Sta­tis­ti­ca. The goal is for AI to act not as a retro­s­pec­ti­ve inter­pre­ter but as an acti­ve par­ti­ci­pant throug­hout the enti­re ana­ly­ti­cal pro­cess, from initi­al ori­en­ta­ti­on through to final con­clu­si­ons.

For manu­fac­tu­ring orga­ni­sa­ti­ons loo­king to get more ana­ly­ti­cal value from their exis­ting data and Sta­tis­ti­ca invest­ments, this is a prac­ti­cal and imme­dia­te­ly usable start­ing point.

++++

The Stat­Soft GPT Con­nec­tor is available as part of Stat­Sof­t’s AI pro­ject offe­ring. Cont­act Stat­Soft to dis­cuss how it can be inte­gra­ted into your ana­ly­ti­cal work­flows.

Categories
Latest News
Your contact

If you have any ques­ti­ons about our pro­ducts or need advice, plea­se do not hesi­ta­te to cont­act us direct­ly.

Tel.: +49 40 22 85 900-0
E-mail: info@statsoft.de

Gui­do Band­holz (Head of Sales)