How to Score with Machine Learning - Acting Instead of Reacting, Part 1
18 Nov 2019 - Industry 4.0, Artificial Intelligence, Production, Technology
Big Data are generated at a high pace and grow day by day. In metal production in particular, large amounts of data from production processes are collected over many years. Finding patterns in these large and complex data sets is impossible for the human brain and even for traditional data processing and management applications. This is where Machine Learning (ML) comes in. In our new Machine Learning (ML) series “Acting Instead of Reacting,” we explain what ML is, how it helps you predict defects in your metal production, and how you can create your own predictive model.
Due to new computer technologies, Machine Learning (ML) is no longer like the ML of the past. Arthur Samuel, one of the pioneers of ML, coined the term in 1959 as a field of study that gives the computer the opportunity to learn by itself without being explicitly programmed. In other words, researchers wanted to see if computers were able to learn from data. The iterative aspect of ML played an important role: the models achieve reliable results based on previous calculations and new data. A new aspect of this concept is the ability to apply complex mathematical calculations automatically, repeatedly and quickly to large amounts of data.
Data are the oil of the 21st century and the raw material with which the ML algorithms are fed: they comprehensively examine all possible combinations of available data in order to find patterns. Especially in the production environment, new concepts of Artificial Intelligence (AI) for ML open up new possibilities for automatic, reliable predictions, for example to avoid errors and optimize business and/or production processes.
In order to understand the complexity of Big Data, here follows a short journey of thought into sports analytics.
A Mind Journey: Will the German National Football Team Win the 2022 World Cup in Qatar?
Imagine that we have the following training data:
Training data
- 13/10/2018 Germany lost against Netherlands
- 16/10/2018 Germany lost against France
If you just look at the results of these two games, can you imagine how Germany played against Russia on 15 November 2018?
Based on this dataset one would predict that Germany had lost because the national team lost all its games before. But it is a wild guess - we do not have enough game data for a more reliable prediction:
Additional training data
- 13/10/2018 Germany lost against Netherlands
- 16/10/2018 Germany lost against France
- 08/06/2018 Germany won against Saudi
- 26/06/2018 Germany won against Sweden
- 23/03/2018 Russia lost against Brazil
- 27/03/2018 Russia lost against France
- 30/05/2018 Russia lost against Austria
If you just look at the results of these games, can you imagine now how Germany played against Russia on 15 November 2018?
With only these data one could guess that Germany had won against Russia, because Russia lost all its games and Germany won some.
So do more data always improve the prediction quality and would thus an increase in the data set for all matches of Germany and Russia since the 1996 European Championships lead to an improvement in the prediction quality?
The extension of the training data by the results since the European Championship in 1996 will not increase the prediction accuracy of the match in 2018, as the teams have changed completely since then.
We need more data features to improve the prediction, such as the score and competition data features of the historical data of the German and Russian games, and information such as coaches, player medical reports, etc. In particular, the score will help to make more balanced and reliable predictions - this is how the team ranking is calculated. In the official FIFA Ranking of November 2018, Germany ranked 16th, while Russia was not even in the top 20.
The combination of recent match history and the FIFA Ranking shows that Germany should have won against Russia on 15 November 2018.
Result: Germany actually defeated Russia 3:0 on November 15, 2018.
Can we now predict whether Germany will win the 2022 World Cup in Qatar on the basis of recent match history, the current FIFA Ranking and other data?
The quality and results of the football team vary over time and cannot be extrapolated too far. The model should be re-trained regularly to ensure good prediction quality.
From Football to Metals
By using algorithms to build models that uncover connections and patterns in production data, metals producers could make better decisions without human intervention. For example, imagine if the software could automatically predict which defects could occur and adjust production processes accordingly in order to avoid them. The models of Machine Learning (ML) provide a solution since they exhaustively explore all possible combinations of how production factors affect quality metrics and defect types. The new PSImetals service for quality prediction also relies on ML in order to be able to predict production-related defects.
What's your opinion on this topic?
Acting Instead of Reacting Series
- Part 1: How to Score with Machine Learning
- Part 2: How to Predict Defects with Machine Learning
- Part 3: In 5 Steps to a Predictive Model
Raffael Binder
Director Marketing PSI Metals GmbH
After taking over the marketing department of PSI Metals in 2015 Raffael Binder immediately positioned the company within the frame of Industry 4.0. So it is no wonder that in our blog he covers such topics as digitalization, KPIs and Artificial Intelligence (AI). Raffael’s interests range from science (fiction) and history to sports and all facets of communication.
+43 732 670 670-61
rbinder@psi.de
- Tags:
- machine learning
- prediction
- metals
