I recently listened to a podcast1 about data analysis in industry. It is in German, and there haven’t been new episodes since two years or so, but the available content is still very valuable. One of the episodes^2 is about data automation in data analysis. I will summarise the content, and I will show strong parallels to automation and digitalization in engineering: a considerable part of engineering work is nothing else but data analysis, and the resons for automation are the same.
Reasons for digitalisation and automation
The world around us is changing. New solutions and technologies allow us to optimise workflows and to increase productivity. Digitalisation and automation are important steps in this process. But why does it make sense to deal with digitalisation and automation?
- Computers do things faster. One of the largest advantages of automation is time saving. A computer can carry out certain tasks much faster than a human.
- Humans make errors. Humans are not perfect, they make errors, and errors can be expensive. A (classical) computer programm on the other hand is very reliable and it will carry out the tasks with high precision, without getting tired or distracted.
- Repetitive tasks are boring. Not only boring, but also tiring. By automating these tasks we can relieve our colleagues and give them the opportunity to focus on more challenging and interesting tasks.
- Growing amounts of data. The amount of data we generate every day is growing. And this data wants to be processed. Digitalisation and automation are important enablers that allow us to cope with growing amounts of data.
- Gaining insights from data. Everybody has heard that data is the new oil and valuable information can be extracted from it. The expectations are high, and tech companies suggest that it’s super easy to gain insights from data usig business intelligence tools, machine learning, or AI. People ask: what can we do with our data?
All these are good arguments for automation. However, digitalisation and automation do not come for free. It costs time and money to collect and clean data, to develop algorithms and tools. The initial investment should pay off after some time. Before starting an automation project, we should figure out how often that task is carried out, by how many people, and how much time (or cost of errors) can be saved. And this needs to be compared to the development cost. If it pays off after some weeks or months, the automation is a good idea.
The engineering view
These were the key point from the podcast episode. Many engineers will think “Hmm, what’s so special about it? We’ve been doing that for decades.” True. Automation of engineering computations is an old topic.
Classical computation
When using classical computation methods in structural analysis, aerodynamics or heat transfer, hardly anybody uses handbooks any more to look up data in tables or diagrams. The formulae have been programmed, curves have been digitised and lookup and interpolation happens in the background. The engineer doesn’t have to deal with manual calculations but can focus on selecting the correct methods and interpreting the results.
If you are lucky then these programs have been written as “proper code”, and operate on open data formats. Because then can be combined, integrated into workflows, run from the commandline and are not hidden in some GUI program with a proprietary data format, or in spreadsheets. Because then you can write shell scripts and run many analyses in batch mode, or easily modify the input and re-run complete analysis process. (DLR tool)
The automated computation has opened the door for optimisation of products. We can do more design iterations in less time. Where in older aircraft structures frames or stringers had a fairly constant cross section, modern structures are optimised way more. It is not uncommon that cross sections change every couple of centimetres, in order to adapt to the local load and result in a very light structure. This was made possible by the automation of computation, and modern, cost-effective manufacturing methods. (DAKOTA)
CAE
We don’t solve differential equations by hand any more - we use FEM or CFD methods to do that. And engineers use CAD tools to document their designs. These programs are all heavy on GUIs, and many people still do stuff manually. However, most of these tools have a macro language or similar, thus allowing automation on typical tasks.
Not-so-typical tasks
Besides these product-related core engineering tasks, there are tasks that may take up most of the time of an engineer. Few people want to know, few people talk about it, and often there is not focus on automation and digitalisation of this kind of work:
- Search
- Information extraction
- Data cleaning, aggregation, visualisation and reporting
- Documentation
A future post will focus more on these aspects of engineering work.