How artificial intelligence and process knowledge helped develop a competitive and efficient system for monitoring small pipes
monitored pipe
Monitoring and detection for each existing stretch
from project to delivery
Setup, model calibration and commissioning
assertiveness
System tested and validated with real leaks
A large mining company had the need to monitor all the piping that transported the ore tailings, from the underflow lines of the thickeners to the dam, totaling approximately 8 km of pipeline.
A team of data scientists, data and software engineers, process and piping experts was assembled to deal with the following challenges:
A system was developed that, using real-time data from existing sensors along the pipeline, makes use of machine learning techniques to model and predict leaks.
The techniques used are known as anolamy detection and, by means of energy balance equations and regressor models, allowed the system to learn the normal behavior of the pipes and any abnormal behavior to be treated as a possible leak.
The system was tested and validated with real leaks, detecting leaks of the order of 70 m³/h in a pipeline that had a nominal flow rate of 1500 m³/h with an assertiveness of 95%.
The rate of false positives was about one per week, or four per month.
Details:
Using Machine Learning, a regression model was developed that predicts an optimal ratio between certain process variables. When this value deviates from the real time measured value above a specific limit, and this difference remains constant for a certain period of time, an anomaly alarm is triggered, signaling a possible leak.
Details:
KNIME is an open source data analysis platform, intuitive and easy to handle. Composed of several blocks with specific functionalities, by connecting and adjusting these simple blocks we can build extremely complex and easy maintenance systems. Thus, with the training and explanations we have done, the client did not need deep programming knowledge to be able to handle and manipulate our models.
Details:
To ensure reliable detection, the system also used a set of algorithms to evaluate the quality and reliability of pressure, flow, density, etc. measurements.
Therefore, the detection models only work when the received data is reliable.
In addition, the system also informs the operation of the plant of the need to check the sensors that present any anomaly, preventing the pipeline from being unprotected and with unreliable measurements for a long time.
The article "Leak detection using machine learning techniques", written by the team based on the work done, received the ABM award for best maintenance and engineering work in 2019.
How artificial intelligence and process knowledge helped develop a competitive and efficient system for monitoring small pipes
monitored pipe
Monitoring and detection for each existing stretch
from project to delivery
Setup, model calibration and commissioning
assertiveness
System tested and validated with real leaks
A large mining company had the need to monitor all the piping that transported the ore tailings, from the underflow lines of the thickeners to the dam, totaling approximately 8 km of pipeline.
A team of data scientists, data and software engineers, process and piping experts was assembled to deal with the following challenges:
We know how difficult it is to find efficient tools and appropriate solutions to your problems. Our team is available to help you choose the best solution, based on our experience in the most diverse areas of the industry.
Article by Leonardo Vieira, responsible for the ihm operation in the US, on how artificial intelligence has changed the use of data in industry.
Article by our director Gustavo Brito about the challenges that the pandemic brings to the industry and what the reality may be from now on.
Matter that addresses the use of automation and data integration in logistics processes, and how this can make a difference.
