banner_tod_all2.png

Industry

Sensor data (IoT), Financial transactions, Plant Information, Detection of anomalies

usine

Monitoring and improving the quality of PI data

Our client, an industrial group with hundreds of subsidiaries around the world, wanted to control and improve the quality of PI data (PI = Plant Information: data emitted by sensors installed on production sites).

The objectives were as follows:

  • Create PI nomenclatures (Assets, Attributes, Tags) with clear naming rules, which are free from duplicates in order to allow better reuse of Tags as well as cross-site analyzes.

  • Set up an efficient monitoring system for PI Tags (= time series): real-time detection of missing or inconsistent data, identification of defective sensors, etc.

  • Supply Data Scientist teams with reliable data, which is an essential prerequisite for building consistent and efficient predictive models (forecasting, predictive maintenance, etc.).

Solution provided by Tale of Data

Harmonization of sensors nomenclature:

Tale of Data automatically reconciles texts (name, description, etc.) with spelling differences using advanced “fuzzy matching” algorithms: phonetics (English / French), consonant (or vowel) frequency , word fragmentation (N-Gram), or even automatic word weighting (Inverse Document Frequency): a low weight is assigned to the least discriminating words.

Monitoring of sensor data using Tale of Data's time series analysis algorithms:

  • Determination, by type of sensor, of the appropriate alert thresholds for the measured values ​​(temperature, pressure, etc.): these thresholds were obtained by launching an automatic analysis over several years of history

  • Determination, by type of sensor, of the appropriate alert thresholds for the elapsed time between two measurements: these thresholds were obtained by launching an automatic analysis over several years of history

  • Setting up automatic alerts if previously determined thresholds are exceeded or when data is missing

Benefits

Labeling harmonization and deduplication have enabled the creation of a shared repository of PI metadata: Assets, Attributes, Tags.

This shared PI metadata repository, with clear naming rules, opened up many possibilities:

  • Consistent representation of the system: same set of attributes for items representing the same type of equipment, with names, descriptions and standardized units of measure

  • Facilitation of "multipoint" analysis: standardized metadata make it possible to aggregate or compare time series, whether it is for monitoring, reporting or predictive analysis (Machine Learning)

 

The time series analysis made it possible to put into production, in a few weeks, a fully automated monitoring system continuously analyzing data from tens of thousands of sensors.

Alerts on very precise conditions have been set up (e.g. sensors emitting erroneous values ​​or presenting anomalies in the time intervals between two measurements). These alerts can be reconfigured by business users at any time, without writing any code.

Other scenarios are possible, do not hesitate to contact us to discuss your business cases.