Data Dashboards - also for engineers

Dashboards have been on the radar of the manufacturing industry for a while. They show up more and more, sometimes connected to the cloud, sometimes integrated with local solutions like a MES or SCADA. In most cases they are used to monitor energy no, production rates or operational effectiveness (OEE).

Using dashboards

Having such dashboards allows operators and management to continuously keep an eye on important production parameters. What they see can lead to action, for (production) process improvement or changes in the setup of production lines.

A simple example could be a check on the amount of material or product that is discarded in a certain production step. Let’s say a product is made of rubber, and the dashboard shows that after running the material through a mold, 12% of the resulting products gets discarded. The dashboard can make this visible in real time, together with the exact batch of rubber that was used, and product in parameters like temperature, air pressure, humidity.

It is possible to show these next to a longer term trend, so that an operator or production mange can see if patterns occur. If so, these can be analysed further to see if changing material, temperature or pressure reduces the percentage of discarded products.

The data needed for this typically comes from digital equipment, like sensors and controllers (PLC), or from human input. The latter for example is applied in metal workshops where the workers indicate on a screen or through a barcode scan when they start and stop working on a certain job. Next to these, we can also add analog signal sensors with a digital output to an existing system, for example to measure current on an analog machine to see if it’s running or idle. This opens up dashboard functionality for older, non-digital equipment.

Using a dashboard as an engineer

The analysis performed as described above leads us to another possible use of dashboards: those that can be used by engineers that build and maintain the production machines. Each of these machine, contains an army of sensors and actuators to control - sticky to the example- pressure, temperature and humidity. Pressure of air or hydraulics controlled by valves, and settj pressure is a closed loop. Sensors mean pressure and as a result of those a controller (e.g. a PLC) change the sells of a value to get the pressure to a pre-configured set point. As an engineer, it could be useful to monitor that process. This allows you to see if pressin variations are small and not to frequent, or if the controller perhaps misses a necessary update. This becomes importent if analysis show that the process issues found in the production dashboard have a technical cause.

Having a technical dash board like this helps reduce analysis time in case of problems in production It reduces the need to go through log files, set traces (or at least less) to find out what is going wrong. Combining the two in one environment allows for even faster analysis. Now this may sound like somthi that already exists, for example in SCADA systems. To some extra it does, but it is definitely not used everywhere. Also it required data to be exposed by controller and tensors so that it can be fed to the dashboard - e. g,through a Unified Name Space. It may also require additional senior to be placed. what if the pressure valve is analog and we can only check it’s status by measuring the current on its control wire? Adding an analog sensor and a A/D converter could make the data available to the world, and the dashboard.

But can AI not do this instead?

Of course, someone will bring up the question whether AI cannot do this, instead of a human with a dashboard. A valid question in times of a hype, especially with a hype that actually shows useful spin offs. However, what most people refer to when talking about AI are ChatGPT and similar systems. These are LLMs, Large Language Models, which learn to respond to questions by training them with a lot of text. These are not so suitable for analysing the data we see here, but other models (like Bayesian networks and reinforcement learning) can certainly be used to construct such analysis systems over time. However, also these are training on historical data and feedback on their responses, so they won’t be functioning ‘from day one’. Also because every production line is in a way unique, and has it’s own specific quircks and charateristics. Having the data collected for dashboard use and then moved to long term storage would allow such applications to be developed over time, but that will not remove the need and added value of dashboards for now.