Predictive maintenance has gained more focus as one of the important application scenarios for the Industrial Internet of Things (IIoT). Roland Berger GMBH, in cooperation with HANNOVER MESSE, conducted a survey of 153 mechanical engineering operation companies that focus on areas such as transmission engineering, mechanical transmission, hydraulic transmission, electrical automation, and robotics. Of the respondents, 81 percent have a predictive maintenance plan, of which nearly 40 percent provide related technologies and services. However, most of the companies are either still in an R&D stage or have not started any related work.
Insights From Predictive Maintenance on the China High-Speed Rail Project
The innovation of China’s high-speed trains has attracted worldwide attention. With the official operation of Renaissance at 350 km/h, China has attained the highest commercial operation speed of high-speed rail in the world. Predictive maintenance is crucial to ensure uptime as well as the security that should be a prerequisite to run at high speeds.
The first step in creating a predictive maintenance solution for high-speed rail is to develop our machine learning algorithms. For this, we needed to acquire large amounts of data for model training. To gather this data in a short period of time, we instrumented eight rotating parts on the high-speed rail test line at the CRRC facility in Qingdao, China. This acted as a proof of concept and provided the valuable data we needed. We measured the vibration and speed of each rotating part, performed screening and analysis on the data we acquired, and ultimately provided insights on the health status of all rotating parts. However, in this prototyping exploration, we encountered five major challenges:
We needed to quickly build a real-time online monitoring system that could synchronously acquire the vibration and speed of 10 rotating parts using one single device, which would require 20 acquisition channels.
To fit the high-speed rail speed, the spindle speed will be as high as 2,200 rpm. And to ensure the frequency range of the samples covering all the fault frequencies, the sampling rate had to reach up to 25.6 kS/s.
We needed to acquire a continuous, high-speed stream of data from multiple sensors for proper analysis.
Multichannel, high-frequency acquisition produces large amounts of data in a short period of time. This data should provide health insights after signal processing and feature extraction, then be passed to the server, which exerts high demand on the computing power and memory allocation of the test system.
The data from all channels had to be both precisely synchronized and time stamped for optimal use during the machine learning algorithm development. This test equipment time stamp had to also sync with the server side so we could build it into the model with data from other sources.
We adopted an NI InsightCM and CompactRIO solution for data acquisition to address the above challenges. We chose a cRIO-9036 controller and NI-9232 sound and vibration input module based on the requirements of channel number and sampling frequency. CompactRIO can fully meet the demand on computing power and memory allocation. With NI InsightCM, we not only achieved the synchronization of multichannel data acquisition, but also maintain the time consistency between device and server by the way of setting the NTP time server.
The team collected high-quality data and ran the predictive health algorithms on CompactRIO, which also met real-time requirements and helped them analyze all the faults in the high-speed rail test with high accuracy.
Integrating Edge Computing With Cloud Computing to Generate Insights From Data
CyberInsight Technology collaborated with NI and CRRC to overcome some of the data challenges associated with this connected, predictive maintenance application. Vibration data is typically sampled at high acquisition rates (approximately 10,000 samples/s to 100,000 samples/s per accelerometer channel). The processing elements on CompactRIO turn this data into calculated features, such as RMS and peak vibration, and transmit them to a server. This saves on bandwidth required between the distributed CompactRIO edge nodes and the NI InsightCM server that is central to the train. Moving this much data around a normal network is challenging, but there are added costs and complexities associated with transmitting this data off a high-speed train. Today, the machine learning research and analytics model creation is happening in the cloud, but the plan for this solution is to deploy a locally distributed network on the train that can calculate features from the high-speed sensor data and perform the diagnostic analytics to help detect early signs of failure. In this planned architecture, the data payload is transmitted by satellite and is more dense with information rather than packed with preprocessed data. This is beneficial as minimizing the data transfer makes the solution more commercially viable in a larger deployment that scales to the number of trains rail operators typically employ.
Expected Business Impact With Predictive Maintenance
Helping engineers connect test and asset data to enterprise networks is an important step in the next revolution of design and manufacturing. CRRC, along with CyberInsight, is using NI technology to connect data from high-speed rail to improve uptime using predictive maintenance. The next generation of CRRC trains should have a much shorter time to ROI because of this exciting technology.
CRRC Qingdao Sifang