SUCCESS CASE #BigData #MachineLearning
Predictive maintenance of Malt Mills through anomaly detection

Damm is a leading company in the brewing sector, with 16 production plants and a presence in more than 133 countries. The company produces and bottles more than 18 million hectoliters per year of different brands of beer, water and other beverages. Damm dedicates around 15% of its net profit to R&D and technological innovation.
One of the key machines in beer production is the Malt Mill. Being able to anticipate malfunctions and plan shutdowns, maintenance and replacements intelligently allows for considerable cost savings and smoother, more predictable production.
Damm is considering the possibility of using machine learning models for anomaly detection in order to estimate and anticipate malfunctions. Malt Mills have different sensors that, together with model information and production characteristics, can be used to anticipate failures. The characteristics of these machines mean that it is not possible to have enough historical failure data to develop predictive models. Most of the historical information is of what can be considered as “normal” behavior, and that is why an anomaly detection approach is chosen.
Keepler developed models for detecting anomalies in the behavior of Malt Mills for one of its production plants. This type of model detects anomalous behavior with respect to historical data. Beyond detecting strong and evident deviations of a specific variable, the models are able to combine minor deviations to understand that, as a whole, there is a behavior out of the norm.
Among the challenges encountered is that the sensors do not always return information for each instant of time (time-stamp) so they must be sufficiently robust models to adapt to circumstances with partial information.
An AWS cloud infrastructure was developed for the reception of data, the different ETL and normalization processes, as well as the training and execution of machine learning models developed in AWS Sagemaker. The results are displayed and monitored in Dashboards implemented in AWS Quicksight.
The main technologies used in the development are the following:
-
AWS Quicksight for the development of Dashboards with results.
-
AWS Athena for querying results using SQL.
-
AWS Sagemaker for the development, training and deployment of machine learning models.
-
AWS Kinesis Data Firehose for streaming data management.
-
AWS Transfer for SFTP to automate data loading from source.
-
AWS S3 (object storage) and AWS DynamoDB (high-performance NoSQL database) storage services.
-
AWS Glue to automate data integration and ETLs.
-
AWS Step Functions for workflow management.
-
AWS CloudWatch for system monitoring.
-
AWS ECR and AWS ECS container management services.
-
AWS SNS for notification management.
-
AWS API Gateway for connecting development with client systems via APIs.
As possible evolutions and improvements, Damm considers the possibility of developing additional models to explain in depth the reason for the anomalies detected. Additionally, it is proposed to integrate the results shown in the Dashboards with alarms and warnings within the production operations.

Keepler is a boutique company of professional technology services specialized in design, construction, deployment and software solutions operations of Big Data and Machine Learning for big clients. They use Agile and Devops methodologies and native services of the public cloud to build sophisticated business applications focused in data and integrated with different sources in batch mode and real time. They have Advanced Consulting Partner level and have a technical workforce with 90% of their professionals certified in AWS. Keepler is currently working for big clients in different markets, such as financing services, industry, energy, telecommunications and media.