Live pneumatic leak detection

[cadcoke5]

I work for a company that makes animatronic figures (i.e. robotic puppets). Our electro-mechanical systems have extensive data on the performance of the device, but our pneumatic systems are lacking in this regard. Actually, most of our systems are pneumatic. We would like to be able to detect things like air leaks, valve failures, and mechanical failures.

We have evaluated using simple pressure switches located by the valves in series to the air lines going to the actuators. This hope was that if the pressure went below a certain value, we would get an "alarm" that there was a significant leak in the system. But, in testing, the back pressure from the 1/8" hose lines was enough that the pressure switch never went low enough to trigger. Also, this system didn't give us feedback if an actuator was not moving as expected.

I know there are systems in industry to run a routine, perhaps the first thing in the morning, to test each actuator line to see if it holds pressure. But, we would like some sort of continuous monitoring of the performance.

Currently our system uses a custom box that accepts a "start the show" command over Ethernet, and then runs an internally stored program to turn off and on the valves it is wired to. It can accept some digital and analog 0-10v signals, and do some internal logic programming. Because of the number of actuators in a small space, we can't put limit switches on the actuators, and the quantity of wires to run would be prohibitive.

I suspect our answer will be some sort of pressure sensor network on the output of each valve, and a flow sensor at the main air feed to each manifold. Such a network would then be continuously monitored, and the live performance data compared to a stored set of data created during the installation. At a certain percentage of data being out of spec, the system would report an error.

I have read about CAN Bus and I2C networking, but have not used either. But, perhaps one of these networks could carry our sensor data, and then a dedicated computer collects all the data to determine if a failure occurred.
Since pneumatic actuators are common in industry, I am surprised that I have not come across an existing solution that is commonly implemented. Perhaps I am a novice trying to invent a solution that has stumped the experienced experts.

I have not come across any discussion about this on the forum. Any thoughts on this?

Joe Dunfee

 

[MikeHalloran]

I'd suggest a pressure transducer on each valve output, linked to an A/D converter (that may not need to be particularly fast or have particularly good resolution).

The deal is, a supervisory program or computer works in parallel with the normal control program. Upon a 'start of show' command, it samples and stores the various pressures with timestamps. At the end of a successful show, you tell the supervisory computer "That's what a good show looks like." Then it goes into a monitor mode, where it samples pressures but does not necessarily store them. It 'alarms' however you choose on deviations from the stored 'good' pressures. I'm not sure that a flow sensor would be of any great value, but you could install one and see if you can make sense of what it's doing. You can help train the detection program by introducing artificial errors one at a time, e.g. leaks in various lines, mechanically restraining a given cylinder, stuff like that.

For further study, look at modern and even less modern vehicular ECUs. They can infer an amazing suite of error conditions without dedicated error transducers. Best example; almost all ECUs can detect a missing, loose or leaking gas cap by manipulating the vapor recovery hardware that they normally control anyway for other purposes. My Z28 detects clogged check valves in the exhaust air injection system, by means that are a complete mystery to me.

Some older ECU code has been at least partially disassembled, and is available online with a little searching. You need a schematic and a port map and a programmer's handbook for a similar microprocessor to make sense of it, but that's all available too.


Mike Halloran
Pembroke Pines, FL, USA

 

[cadcoke5]

Mike Halloran said; " I'm not sure that a flow sensor would be of any great value."

My thinking is that an actuator that is not moving because of a mechanical blockage will show the appropriate pressure, but it will do it without any flow. Though, I suppose this might be able to be deduced by monitoring the pressure drop that occurs as the actuator is moving. Though, if the actuator is blocked mid-stroke, that may not work. Also, in situations where an actuator is bearing weight, there may not be much of a pressure drop.

I think the biggest strength of the flow sensor is that it immediately gives me information about the failure most likely to occur; a cut line.

Another issue is cost. A simple pressure threshold switch seems to be the cheapest, and easiest to monitor. We have located some at $12/ea. But, we would obviously loose some of the benefits of more detailed analog monitoring. And since these $12 versions have a pre-set trigger point, we would also need to experiment to know what pressure to use as the trigger point.

I have come across some Honeywell pressure sensor chips in a DIP package for under $9 at Mouser. They have a barbed fitting, and also need a regulated 12v power supply. I will look more in to these.

Thanks for the reply,

Joe Dunfee

 

[cadcoke5]

We ended up simply using a pressure switch, and experimented with where we placed it. No placement was ideal in all situations, so rather than make the system more complicated, we simply accepted its limitations.

For example, we put a pressure switch after a simple needle valve (i.e. without an integral directional valve). this went into a line that was actuating a cylinder that was to move slowly under a light load. The pressure switch would not show pressure until after the cylinder had reached the end of its stroke. If we put the same switch between the needle valve and the directional valve, we instantly detect pressure when the valve opens. But, it could not detect low pressure if the line were disconnected at the cylinder and air were flowing through the line. So, in this case, we felt that putting the pressure sensor switch after the needle valve would be a better choice.

Our software will be written to look for pressure at the correct times for when our pressure switches should activate, and activate the error line if it detects a discrepancy.

I also looked to see if there might be a product that would be part of a manifold assembly, and have pressure sensors in each valve. I imagined each valve might have a sensor designed to work on a CAN bus, so you could keep adding more valves to the stack, and the pressure readings would be passed along the CAN bus. But, I couldn't find anything like this.

-Joe Dunfee