Motivation

Many manufacturing lines currently lack a digital twin that can accurately predict the outcome based on input parameters. As a result, fitting parameters again requires a time-consuming trial and error process. In the case of glass manufacturing, conventional stochastic models have proven inadequate in predicting outcomes.
However, through a research project with our well known partners, we developed a groundbreaking AI system that can accurately forecast manufacturing line outcomes solely based on input parameters.

Our Aim

Our aim is to serve you with an End-to-End deep learning model that is able to predict the parameters your manufacturing line based only on the desired shape you want to manufacture.
After the model is trained through our software you will be able to simulate your manufacturing line accurately. You will be able to run simulations with any product of your choice.
Through that manufacturing experts will be able to focus on tasks that really require their expertise.

HSA TRANSFORM

One of the most common problems in machine learning projects is to define a target specification which can be understood by the model.
With our HSA Transform tool we solve this problem as our software can easily transform various dataformats into a target format that can be handled by our machine learning model.
Our HSA TRANSFORM uses data pipelines in a unique plug and play way that will make you able to inspect a wide range of features.
Furthermore through these pipelines we are able to provide a technology that can teach your model to understand the complex correlations between your input and output data.
In the figure below yo can see an example translation from three dimensional to a two dimensional space.

As you can see we managed to translate a 3d shape from a CAD file to 2d curve which could than be compared to the surface measurement of a real workpiece.
Through translations like in the figure above, you will be able to receive more precise quality values than ever before!

HSA COMPARE

After transforming your data into a usable format. You will want to know how good your manufacturing piece really is. Here is where HSA COMPARE comes in. Within HSA COMPARE you can select the type of error value you want to obtain for your workpiece, wheter it is a scalar value or a error value distribution. This all possible with just a few clicks.
In the picture below you can see the error distribution of a workpiece which was calculated by our COMPARE software.

With HSA COMPARE you will be able to inspect the error of your manufacturing piece and in the same time you are creating a ready to use dataset to train your digital twin!

HSA INSPECT

Our development has resulted in a powerful tool called HSA INSPECT which gives you hints on how to optimize your manufacturing parameters. HSA INSPECT uses mainly explainable AI tools that will help you understand which parameters really influence the quality of your products.
For example the picture below shows a bar chart of the parameter importancy analysis of one of our clients machine parameters. The analysis with our HSA INSPECT resulted that out of the about 40 parameters which were used for optimization only about 10 parameters had noticeable effect on the manufacturing quality of the product.

Process Monitoring

HSA allows the continuous monitoring of process (simulation) data and can extract relevant characteristics from it.

The image below shows the progression of the peak-to-valley (PTV) and root-mean-squared (RMS) values of a target product surface during its forming process; both are monitored for quality control.

View of the current state (increment – blue) compared with the target state (mold – orange)
View of the distance of current and target state along z-axis.
Progression of quality values peak-to-valley (PTV, red) and root-mean-square (RMS, blue) throughout the process.

These values cannot be measured directly and must thus be predicted using the machine settings as input data.

HSA then uses a combination of simulation and real-time machine data to predict the resulting quality outcome.

This then is fed back to a programmable logic controller (PLC) that adjusts the machine settings for a favorable quality result.