Lubella is a well-known brand that produces pasta, cereals, groats, flours and snacks. In 2003, the company became part of the Maspex Wadowice Group – one of the largest companies in the food market in Poland, and Central and Eastern Europe. In 2020, BRAINHINT, together with Siemens, implemented an anomaly detection system for a pasta drying room at the Lubella production plant. Extensive research enabled the prediction and prevention of various unfavourable events in the production process, which in turn is expected to contribute to cost reduction and improvement of the Lublin company’s competitiveness.  

The pasta production process requires a rigorous regime – even the smallest irregularities disturbing it may result in the entire batch failing to meet the precisely defined quality standards and requirements.

Artificial intelligence helps to ensure consistent quality

The Lublin-based company sought to develop an analytical system that could provide continuous monitoring of the pasta production process and predict any possible irregularities during its drying process. Consequently, it would let the Lubella plant reduce the volume of production waste. Another important aspect of using such an analytical system was to optimise the use of plant utilities – benefiting, in particular, from a cost reduction for electricity and water consumption. Other expected benefits would cover an enhanced decision-making process increasing operators’ accuracy and resulting in an overall higher quality of the produced pasta.  

At the beginning of 2020, Lubella decided to implement the project based on the analytical engine provided by BRAINHINT. The purpose of this implementation was to improve the operations of the production line with the use of artificial intelligence. The project of the advanced data analysis system aimed at ensuring more efficient, effective and sustainable industrial production and was carried out in cooperation with Siemens and with the full support of Mirosław Pastuszak – Lubella’s Technical Director and their Innovation Management Department.  

At the turn of May and June 2020, the collection of the first production data began to feed the system, and the first analyses were carried out to create a preliminary anomaly prediction model. In the following months, the model was refined and e-mail notifications for production anomalies got implemented.  

The implementation process took over 900 hours of analytical and development work. The solution is based on the algorithm that uses autoencoders based on LSTM (Long Short-Term Memory) neural networks. To predict the outcomes of the production line, a range of data collected by sensors is transmitted to the Siemens IPC controller, providing insights into statistics such as humidity, temperature, information about the ingredients dosing ratio (eggs and water), and the duration of the pasta drying time. The analytical system model is reflective of the behaviour of the actual production line and informs about likely occurrences of an anomaly with 80% accuracy. This means that the neural network can assess deviations across the production processes and predict the possibility of failure, achieving eighty per cent efficiency.  

How does the prediction system work?

Several key elements can be distinguished in the architecture of the analytical system created by BRAINHINT: PLC controller, IPC, SCADA system and graphic interfaces providing the output data for the supervisor and the operator.  

Sensors from the production line collect, among others, information on temperature and humidity. This data goes to the SCADA system, which aggregates and monitors production information. The aggregated data is then sent to the analytical system running on the Linux server. The output data is transferred to the operator’s device via the graphic interface, combined with audio-visual notifications. Prediction data can also be displayed directly within the SCADA system. In turn, the supervisor’s system is provided with historical data, information on the system sensitivity configuration and model parameterization.  

The key information obtained through this architecture covers e-mail notifications about anomalies. Each time the model results indicate an unusual event somewhere across the production line, an alarm is triggered and a relevant notification is sent to the operator’s inbox. Ultimately, notifications about failures will also be displayed through the dryer operators’ dashboards. Comprehensive reports summarising the production data following each alarm are also delivered 15 minutes after the anomaly occurs. These contain graphs of humidity, temperature and ingredients dosing process, covering the period of 15 minutes before and after the alarm. They help the operator discern better whether the alarm generated was valid.  

Other important functions of the system include the possibility of cross-checking the batch quality. There is a summary of events generated when the batch that triggers the alarm leaves the dryer, and the chart shows the conditions under which that given portion of pasta has been dried. Any occurrence of unusual behaviour is marked in yellow.  

Thanks to the data supplied by the system, it will also be possible to analyse the conditions in which a specific portion of pasta is dried. For easy navigation and to follow the full course of the production conditions, the operator can scroll the screen with a slider. The analytical data is also used to control the quality of final products more efficiently. Further work on this system will enable the development of this already proven solution to be implemented across a wider industrial setting. 



Contact us

We will get back to you within 24 hours.