Taking industrial process efficiency to the next level by high-quality data analytics and process modelling
Data analysis is a great tool in ensuring regularities in an industrial process. High-quality data analysis combined with deep process understanding is a combination that provides a strong basis for increasing the efficiency of the process.
When developing digital twins of industrial operations, it is necessary to understand also the invisible parts of the processes, react fast to changes and automate at least some part of the decision making. It is hard to see the status of the system by observing one process, especially if the reactions are delayed and variables cross-correlated. Sometimes the desired solution can’t be reached by the limitations of the other process parts or some negative synergies. This requires digging deeper into the data by multivariate statistical analysis, model assessing and optimization.
Industrial processes consist of systems that interact between each other. The data of these systems can be signals coming from a device or software, data from cash or material flows as well as an action log recorded by operators.
Usually when applying simulations, machine learning or AI in industrial operations, the key part is model selection. However, if we are talking about large systems consisting of multiple variables, these models are usually black boxes and require analysing the data only based on their input and output data. If we are talking about time-dependent processes, reactions can experience lags, which makes finding correlations more complex, and some changes might show up as random variation in the data even though they are not.
Complex structure and noise, both common in big datasets, cause difficulties in processing and require a lot of computing and powerful algorithms. There might be situations where no prior information is available, and the typical black box approach must be used in the analysis. This is also where mathematics and engineering expertise are linked together.
Get the benefits of the digital twin of the industrial processes
There are clear steps required before the benefits of the digital twin of the process are available:
- Restructure the data:
We start the analysis by restructuring the data so that it is suitable for the analysis method itself. This usually requires most work, as values or other error-causing anomalies might be missing. During this step, the analyst also familiarizes herself with the data.
- Choose analysis technique:
After preprocessing, we choose a suitable analysis technique to use in the specific process. If we are talking about multivariate time series data, for example, we might want to analyse correlations with some vector autoregressive model, principal component analysis etc. With correct analysis, one can estimate the parameters crucial for the analysed system, spot exceptional behaviour and create a basis for a follow-up.
- Create a process model:
Use the estimated parameters from the analysis to create a process model, so that the relevant variables can be forecasted or re-adjusted towards the desired outcome.
AFRY combines digital solutions with deep process technology know-how
In AFRY Smart site, the emphasis of data analytics is on the industrial operations in project sites and clients that aim to
- develop their processes through estimating outcomes and progress KPI’s
- improve communication and decision making by better reporting
- understand their data and the capabilities of data usage
We utilise modern techniques from predictive control and forecasting to machine learning and AI solutions, and we also have a strong understanding of applicability to the different industrial processes.