Shipboard Machine Monitoring for Predictive Maintenance

In 2004, BP became aware of a wireless networking technology called motes, 2 X 2 in. devices that combine a processor, solid-state memory, a radio, and an I/O board that can interface with sensors. These motes could potentially be used to tie sensors into a wireless network, which then could pass data to back-end systems. But most of the motes were going to academics, laboratories, and industrial manufacturers that were still evaluating the technology. No one was actually creating commercial products in those days.

"Once we found this technology, we had to determine if it worked in our industrial environments," says Harry Cassar, BP's technology director, digital and communications technology. "The motes that we saw were delicate, almost bare printed circuit boards. We wondered if they could be put into environments with high metal content, such as a refinery, where there are also high levels of vibration and extreme temperatures. Could they be packaged in a way that would allow them to work in these environments? The only way to find that out was to try it. And that is how the Loch Rannoch project was born."

The Project

The Loch Rannoch project began with the testing and evaluation of mote technology, but BP didn't stop there. The project was actually a multiphase effort that went on to develop a commercial wireless sensor networking system that could be used in BP's industrial production facilities around the world. Each phase encountered its own hurdles and produced its own solutions.

BP's first hurdle was to decide where to conduct its project. Most of its operations required the use of intrinsically safe equipment. To test the wireless networking technology, which did not have that certification, the company looked for an environment that had some of the harsh conditions found in its facilities but that did not require intrinsically safe equipment. The company settled on the Loch Rannoch, a 1000 ft., 132,000 ton oil tanker that shuttles oil from a storage vessel called the Schiehallion to an oil-processing terminal at Sullom Voe, in the Shetland Islands.

The ship had a higher metal content than a refinery, making it a tougher environment from an RF point of view. In addition, the ship had compartments that could be shut off by watertight doors, so there were steel barriers between one compartment and the next. There was also significant vibration from the main engines, generators, and thrusters, and the temperature inside the engine room was between 80°F and 100°F.

BP's chief technology office worked with BP shipping to decide on the focus of its pilot project. "If we were going to do a test on the Loch Rannoch, we suggested that perhaps it would be good to choose an application that would have some real benefit rather than just doing a test to see if the motes could talk to each other," says Cassar. "And BP shipping said that they would like to collect vibration data from a set of rotating machines in the engine room."

The System

BP's project developed a new predictive maintenance system capable of monitoring critical rotating machinery, such as the pumps and motors in the Loch Rannoch's starboard engine room, using vibration data to evaluate operating conditions and wireless communications to send alerts when wear and tear was detected.

"Vibration data can help you understand how a machine is wearing and help predict when you should do maintenance," says Cassar. "Condition monitoring can tell you when a shaft is out of true or when a machine is out of balance. It helps you move from time-based maintenance—stripping the machine down, say, every 500 hours—to doing maintenance when it is required."

The condition monitoring was based on measurements from 150 Rockwell Automation accelerometers mounted on the machinery. Each machine was instrumented with as many as six accelerometers—one for each axis and at two additional measuring points—and a tachometer that determined how fast the machine was running and provided the angle of the phase. The sensors were hard-wired to Intel motes, which were mounted in a metal enclosure roughly 2 ft. from the machine. Each mote can accommodate as many as 10 sensors at one time, although BP didn't fully populate its motes. The mote's I/O board converted the sensors' analog signals to digital data.

Clusters of motes created subnets that were controlled by an Intel gateway. The motes transmitted the newly converted digital data to a gateway using low-power radios built to the IEEE 802.15.4 standard and a direct-sequence, spread-spectrum modulation technique. The motes on the Loch Rannoch were tuned to the 868 MHz range to comply with U.K. standards. However, the radios are tunable, which is important if they are used internationally.

 Figure 1. Loch Rannoch wireless sensor network
Figure 1. Loch Rannoch wireless sensor network

The gateways collected the mote clusters' data—much like WiFi access points— and communicated with other gateways via IEEE 802.11 radios using a mesh network architecture, ultimately passing the data to a controller gateway, which sent the data to back-end data loggers. Each gateway had 2 GB of memory, so it could continue to collect data from the motes even if it was temporarily out of contact with the ship's network (Figure 1).

Problems and Solutions

The Multipath Effect. During the Loch Rannoch project, BP, Crossbow, and Intel encountered a number of problems that had to be solved before the implementation could succeed. The metal structure of the ship and the closely situated machinery made the engine room one of the worst possible environments for RF communications (Figure 2). "Metal is a great reflector of RF energy," says John Suh, a senior applications engineer for Crossbow Technology Inc. "This causes a multipath effect, which greatly distorts the radio signal. So you don't have a good chance of getting the information from the transmitter to the receiver."

Figure 2. The project's predictive maintenance system monitored the pumps and motors in the ship's  starboard engine room. Nearly 150 accelerometers attached to the machinery provided vibration data to   evaluate operating conditions, and wireless motes and gateways delivered the wireless communications used  to send alerts when wear and tear was detected
Figure 2. The project's predictive maintenance system monitored the pumps and motors in the ship's starboard engine room. Nearly 150 accelerometers attached to the machinery provided vibration data to evaluate operating conditions, and wireless motes and gateways delivered the wireless communications used to send alerts when wear and tear was detected

The solution to this problem was the mesh networking architecture. In this type of network, each node communicates with the nearest neighboring node. The data are relayed from neighbor to neighbor until the data get back to the controller or collection point. This architecture is reliable because of the redundancy of paths on which information travels. In the event that one node fails, all the other nodes can still communicate with one another, directly or through one or more intermediate nodes.

"Unlike traditional wireless systems, where you have point-to-point connections via a transmitter and receiver, the mesh network that we used does a lot of diverse routing," says BP's Cassar. "The nodes cooperate with one another to see around corners. They can get around objects and issues that typical point-to-point systems cannot. Mesh networks tend to go around those problems and are more resilient."

Intrinsically Safe Devices. Another operating environment-related problem called for the development of intrinsically safe enclosures for both the motes and batteries. Here, Crossbow's expertise came to bear again, resolving the problem during the phase when the project team was developing the commercial system that was to be used in BP's facilities. "We learned better packaging techniques and designed a special enclosure that would be suitable for applications and environments that required an intrinsically safe rating," says Suh.

What To Do with All Those Data. In the final phase, the project encountered a problem that revolved around vibration analysis' need for large amounts of data to accurately identify mechanical problems in machinery. The challenge here was to reconcile the high volume of data generated by the sensors monitoring the machines with the low data rate protocol (IEEE 802.15.4) used by the motes to communicate with the gateways.

The solution was to reduce the amount of data that had to be transmitted to the gateways by performing the bulk of the data processing on the motes. Initially, the motes had 8-bit processors, but these proved inadequate for the amount of processing called for. So Intel and Crossbow substituted Intel's 32-bit X-Scale processor.

"When you look at sensor network applications, there is a range of data bandwidth requirements for particular sensors," says Crossbow's Suh. "In some cases, you may sample temperature with a simple thermistor or RTD, and that doesn't take much processing power to get the data. But with this application, we are using higher-bandwidth sensors—vibration sensors. We decided that we needed something more powerful than the 8-bit processor, so we went to a 32-bit microprocessor."

Intel further enhanced the processing power of its motes with software. "Typically, you do an FFT and other calculations on the data to extract the frequency characteristics," says Ralph Kling, director of sensor network operations for Intel. "We licensed some of Rockwell's IP in that space to do some of the more advanced processing, and we ported the algorithms from Rockwell onto the motes to benefit from their experience in terms of doing this kind of monitoring. Our goal was to do a lot of the processing at the edge of the network, instead of in back-end servers, as it was previously done."

Intel also added software to the gateways to perform XML formatting of the data. "We have an XML-based system that allows data abstraction and querying of the data," says Kling. "XML is a generic intermediate format that we use for exchanging data between the gateways and the back-end. The gateway software also provides an interface on the back-end that talks to a Rockwell server and produces the data in a format that the server can work with. The data can be visualized and processed with the existing equipment on the ship."

More Power to 'Em

As with any wireless sensor networking application, power was a big concern when BP developed its commercial system—how to get power, how to use it, and how to ensure there would be enough. To address these concerns, the project team covered the basics by using low-power radios built to the 802.15.4 standard, as well as implementing power-management circuitry on the motes.

In addition, the work done to reduce the volume of data transmitted, and thus the number of transmissions, also resulted in power conservation. "We reduced the number of radio transmissions," says Cassar. "The radio is the component that uses most of the battery life, with the processor surprisingly taking quite a small amount. This increased battery life dramatically."

Another piece of the puzzle was power harvesting. "One of the technologies that we tested on the Loch Rannoch was energy harvesting that takes the vibration from a machine and converts the kinetic energy to electrical energy," says Cassar. "We have two forms on the Loch Rannoch. One was charging lithium ion batteries to see if we could fully operate a mote. And the second was charging super capacitors as the storage medium. We understand that energy harvesting will play a big role in motes."

Proving It Out

The Loch Rannoch project produced an efficient automated data collection system for machine monitoring and predictive maintenance that eliminated many of the manual processes used in the past. "Condition monitoring data from many types of rotating machines was typically captured by operators using handheld devices, resembling PDAs, checking one accelerometer at a time," says Cassar. "At each machine, the operator recorded the location at which the measurement was taken and captured a few seconds of data. It was a very tedious job, and they got around to doing it only every six to eight weeks—sometimes even longer. The amount of data that they picked up was actually quite small because they could collect the data only a few times a year. We had our network set up to sample every 18 hours. So instead of doing it every eight weeks with the handheld device, we were doing it automatically every 18 hours."

Other lessons were learned in the Loch Rannoch project:

  • 1. Sensor networks work in hostile environments. In fact, they work well.
  • 2. The choices of radio and network architecture are important. Both increase the reliability of communications.
  • 3. Advanced platforms are a good match for this type of application because they dramatically reduce the volume of data to be transmitted with local processing at the edge of the network.

Tom Kevan is a freelance writer specializing in technology. He can be reached at [email protected].