Intelligent Predictive Maintenance System Based on Container Technology, IPM_c

Predictive Maintenance (PdM) can predict when a machine needs maintenance by building a model so that it can be planned in advance to ensure the normal operation of the production line and avoid the unscheduled shutdowns. The Remaining Useful Life (RUL) prediction algorithm of the original IPM system is based on an exponential model, which has shortcomings and leads to inaccurate RUL prediction. The Time Series Prediction Scheme (TSP Scheme) is used to replace the RUL algorithm of the original IPM system to make the IPM system more accurate in application. The Intelligent Predictive Maintenance (IPM) developed by iMRC team overcomes these shortcomings by using the Time Series Prediction Scheme (TSP) and leverages the advantages of containerization technology for rapid deployment, fast failover, and light weight, and along with cloud computing, it can be quickly deployed to all production machines and been managed easily to ensure the production quality.

To avoid unscheduled shutdown of production machine, proper maintenance is required. Regular inspection and maintenance are a simple method, which is also known as Preventive Maintenance (PM). However, this method is not cost effective. If the frequency of periodic inspection is set too intensive, it will result in waste of manpower and loss of machine capacity. On the other hand, if the inspection interval is too long, it may lead to unpredictable crashes. Predictive Maintenance (PdM) can build a model to predict when the machine needs maintenance so that it can be planned in advance to ensure the normal operation of the production line and avoid the unscheduled shutdowns. This can also improve process capability and utilization rate at the same time.

The Intelligent Predictive Maintenance (IPM) system developed by Fan-Tien Cheng's research team, shown in Figure 1, is a general-purpose technology that can be applied to various industries. The system has been successfully applied to the semiconductor packaging, LCD panel, solar energy, wheel machining, and bottle blowing machine industries. It can predict the future abnormalities of critical components before they fail, so that machine maintenance technicians can take necessary preventive measures in advance to effectively improve the performance, efficiency, and availability of the automation equipment.

Figure 1 Intelligent Predictive Maintenance (IPM) System

The IPM system architecture consists of an IPM modeling server, an IPM management server, and multiple IPM servers, including a  Cyber Physical Agent (CPA) with a predictive maintenance module embedded in the CPA. The IPM system can monitor the health status of multiple targets throughout the plant. For example, if a production machine in a factory has multiple target devices, to monitor the health status of all the target devices in the production machine, multiple CPAs in an IPM server can be assigned to retrieve the sensor data of the target devices and monitor the multiple target devices; then, the predictive maintenance module in each CPA can be applied to monitor the health status and remaining life prediction. Finally, the system also provides factory-wide management and equipment interfaces for users to view the health status and remaining life of the monitored machines.

However, the original Remaining Useful Life (RUL) algorithm of the IPM system is based on an exponential model, which has shortcomings and leads to inaccurate RUL prediction. The Time Series Prediction Scheme (TSP Scheme) is used to replace the RUL algorithm in the maintenance module of the original IPM system to make the IPM system more accurate in application. The iMRC team also leverages the advantages of containerization technology for rapid deployment, fast failover, and light weight to refine the IPM system architecture with cloud computing, so that it can be quickly deployed to the whole production line in parallel and been effectively managed, and the production quality can be ensured.

The IPM_C system with time series predictive algorithm can solve the problems encountered in traditional predictive maintenance.

(1) The time series model built by applying information criterion can solve the problem of model inaccuracy caused by sudden crashes and sudden changes in estimated trends that cannot be prevented by the exponential prediction algorithm applied in general predictive maintenance.

(2) The IPM_C system is proven to accurately predict the remaining life of the machine and inform the user when the machine will fail in the future and whether the machine is aging or not, so as to maximize the normal operation time of the machine through the remaining life estimation.

(3) Provide Pre-Alert Mechanism (PreAM) to avoid no pre-warning downtime, reduce the loss of production line downtime through pre-warning lights, and reduce maintenance costs and maximize production time to reduce maintenance budget; and

(4) Provide the Death Correlation Index (DCI) to present the similarity between machine and failure mode by predicting the change of the death value and the current moment, and estimate the possibility of machine entering into the death state.

Innolux Corporation and Chum Power Machinery have successfully implemented the IPM_C system. In Innolux, the health status monitoring of 16 Turbo pumps in the dry etching machine has been successfully implemented. We will carry on the health status monitoring of approximately 170 Turbo pumps in the future of the whole factory. In Chum Power, we have implemented the system to 5 different key components of the bottle blowing machine and completed the parallel expansion process. The overall return on investment of 363% and a payback period of 7.8 months.