background

MACHINE PREDICTIVE MAINTENANCE

Covid-19


MACHINE PREDICTIVE MAINTENANCE

Middle Eastern Heavy Industry Polymer Manufacturing company leveraged Predictive Machine Maintenance to create a comparatively cost-effective production process to forecast the machine failure prediction using telemetry data.



  • Challenges : This heavy industry polymer manufacturing company wanted to increase the cost effectiveness and the overall efficiency in the production department. Due to mechanical problems in heavy industrial plants, delays in efficient processes are expected to be forecasted and resolved. The business issue is to predict that a system will malfunction because of a certain component failure in the near future irrespective of the telemetry data available.

  • Approach : While developing this application we at Aximind took data from the industry which helped us to stack the accessible data which is required to forecast the machine failure. We used the IOT Framework and Data management process to extract the accessible data from the unstructured data collected from the industry. The data we used are mentioned below:
    ● Failure History & maintenance history (Component replacement records)
    ● Machine Conditions and usage
    ● Machine & Operator features
    ● Telemetry Time Series Data (Voltage, Rotation, Pressure & Vibration)
    ● Error Data (Non-breaking hours)
    ● Machines Schedules
    Post preparing the model, we validated the same with the client data which was provided to us to train our model for accuracy check. The model along with a secured ML UI was deployed at the client side plant as a Real-time Condition based Monitoring System.

  • Solution : At Aximind, we had developed a predictive maintenance application based on feature engineering to put together all the various data sources to generate features that better define a system physical malfunction in real time. Time series data helped measure the lag difference and we used the LSTM algorithm to construct the model to find the likelihood of system failure Time Series data. In addition to these methods, Aximind also used a time dependent division technique to divide the time based on the desired dimension and Gradient Boosting classification to improve the model accuracy to predict the failure of the system. At the end final models were deployed in Hadoop and were run on a daily basis for almost real-time predictions every 45 minutes.
  • Impact :
    ● The operators and plant supervisors were given forecasts every 45 minutes for one hour to advise the support team about taking suitable actions
    ● The predicted machine failure yield improvement was around 0.045% based on the client manufacturing data and machine information at the factory location
    ● The annual savings of about $150K in the given client factory have been observed and the overall productivity has increased and time period of manufacturing have been decreased significantly

Do you want to work with us?

Developing a plan that is custom-built for your business.