Industrial video surveillance: we remove technological data from cameras and predict emergency situations
Very far beyond the Moscow Ring Road, where only geologists and rare bears roam (looking ahead - they are also addicts), there is an industrial complex. Mined ore arrives at the entrance, and concentrate at the output, which can later rot.
If under certain conditions a worker enters this zone, his family will be paid insurance
At this plant, of course, installed a bunch of sensors that report accidents. However, a second confirmation loop was needed, because before the start of the project, a person necessarily resorted to the shop or point and visually confirmed the situation. Before that, it was impossible to take any critical measures.
')
This is a color frame from one of the cameras. Everything, in general, everything is covered with gray dust.
In contrast to the warm and attractive offices and sterile data centers, on this site constantly sharashit electromagnetic radiation from cables with a leg in thickness, interference from start-ups of different equipment. And then there is a great dust that accumulates centimeters per day.
So, there is a set of sensors at the factory that signals some abnormal situations to the dispatcher. Since the sensors very often behave strangely, plus they give a far from complete picture of the scale of the situation and the nature of its development in general, in order to take a serious decision, you need to send a person to see what is happening. An intercom dispatcher calls the supervisor, and he is already looking for a brigadier and asking to be sorted out. There are two answers: “No, the computer drives” and, less often, “Mikhalych, it's ####, let's go #### urgent!”. A few minutes later comes a verbal description of the accident. If the accident is serious - in a very concise, capacious and figurative form. It is this form (delayed by 4-5 minutes) that causes the stop of the conveyor, emergency pressure relief on a number of units, breakage of expensive equipment during an emergency shutdown, and other things, after which the company usually suffers financial losses.
With financial losses, of course, it was impossible to put up.
Initially, the customer turned to us to calculate the cost of the machine vision system, which allows to automatically monitor all problem points. A list of nodes and a list of possible accidents was attached. For example, in this vat nothing should overflow; this node should not work as long as people are nearby; out of curiosity, bears sometimes come here to the pipeline; here there is a gap in the pipeline, and here it is necessary to ensure that a large piece of garbage, garbage — or tape does not break — does not go along the conveyor.
Looking ahead a bit, I'll show you a frame from the line where, for example, you need to control passing crowbars, sticks and another large-sized one. In the west, they still control the size of the fraction, but this is already much more expensive because of the requirements for the picture, the lighting, and because of the features of the code:
For such systems, there are ready-made industrial solutions developed for American, German and Japanese productions. Their main problem is that each non-standard detector requires a serious amount of code and tests for recognition. We talked with both domestic and Western developers of similar systems, and considered the first budget. The main problems are these non-standard detectors and plots that are difficult to recognize as above with almost no contrast.
The budget was very large, so the customer left to think further. According to his calculations, it turned out that it is much cheaper to raise a control tower in the 10-year perspective and put people who know the production like the back of their hand and are perfectly able to recognize abnormal situations. Plus, tighten the basic automation like the focus on the picture for triggering the sensor, basic motion tracking in areas where there should be no movement, etc. In general, a person should be - like a grandmother on an escalator in the subway.
So, we assembled the system on the basis of cameras, simple sensors and, in the program part, basic (library) detectors.
Correctly the system is called "technological vision". The technologist remotely looks at what is happening there at the plant and does not enter the aggressive environment. But in this aggressive environment of the plant are our cameras. The concept of technological vision still includes such solutions as specialized analytics on a node. For example, there is such a machine, the task is to mix the components and get a certain type of foam at the exit. Machine vision determines the size of the cell foam - depending on the size of the cell, the analyst determines which reagent to add and immediately gives a command to the machine controller. We have not yet implemented this, but, in general, it is quite real. Another thing is that, I repeat, it is here that, due to the special conditions, a living technologist is more economical.
The first problem of the project was that it was necessary to make a specific project proposal. If on a regular project it can be done in a week of sitting with plans and models and one departure, then it did not work out this way. We went there several times. We walked around each shop, talked with technologists, asked everyone what they needed to see and how they wanted the sensors to be located.
Here is a picture on the phone with a survey and a delineated field of view of the camera
The dimensions of the plant are huge (tens of kilometers), there are not always precise layouts, there is no possibility to attach to something on site to make a plan for installers in the future. As a result, we simply removed certain places and marked points for the cameras and another weak point on them. Like that:
The point is to remove not only the monazh point, but immediately together with the technologist what the camera will see. To do without neponyatki.
We did this for about two weeks. Typically, the design goes so much, but it took 2 months to complete the project. I must say that I walked around the facility with a service notebook, which during this time fell a couple of times, crashed, and even six months after putting the system on it, there were still traces of local dust.
Switching itself was relatively simple. The complex contains a ring of optics with branches, so there are no problems with the channel for the HD picture. Optics are there, by the way, because copper simply will not work. Long crossings between workshops are possible only on existing cable racks. And there are power cables with a cross section of the core in several centimeters. And there are not a dozen of them living there. Copper was only from shields to the cameras themselves, and in the cable with several levels of shielding.
The shields were assembled by themselves - they calculated the temperature regimes and put industrial noise-proof switches inside. As practice has shown, it works great, and in no way becomes a bottleneck. The problem was that there are no cameras as such for an industrial very hard use. In the sense that they still have the appropriate certificates. Therefore, again, it was necessary to take in special housings, to make grounding, to fasten special "wipers" from above to shake off the dust. Housings with wipers were found in one large vendor very well and immediately got up as it should.
Of the 165 cameras on commissioning, 19 fell off immediately. This transportation or factory defect is unknown. It was necessary to check in Moscow after arrival, but the customer needed as quickly as possible, so we sent immediately to the complex without unpacking.
When mounted their tracks, we learned what legacy-isolation is. There are many lines at the plant that have not been used since the times of the USSR, because they are duplicated, replaced and not unmounted, etc. In most cases, the customer gave us trays, and we mounted them. But on a number of runs there were consoles with large distances between the supports. It felt great thick self-supporting cable, and everything else was attached, in fact, on it. So, after 20 years of operation in an aggressive environment, the insulation on such cables turns into dust. While standing and watching - everything is fine. You touch the braid - it is poured with a quiet rustle.
In one of the legacy workshops
Previously, a pipeline breakthrough could erode an embankment at a nearby lake in 15–20 minutes. Now the blur time has not changed, but the reaction time has changed. Visual confirmation can be obtained immediately and immediately make a decision. Previously, this required to jump into an all-terrain vehicle and go to the other end of the territory, for example. There is still need to add that it is not scary on this lake from the complex to walk only during both weeks of summer. All the rest of the time there are hot parties for addicts addicts. These animals adore strong unusual smells - they are like bears in Habr's reading at IT people. Therefore, checking the sensor every time was somehow not very interesting. We were also told stories about wolves, but we did not see them.
In addition to reducing the reaction time, the length of downtime after the accident also decreased. Previously, a big problem after a stop was to check that all repairmen left the site. This was done along the same chain and with difficulty. Now there is still a visual confirmation.
If under certain conditions a worker enters this zone, his family will be paid insurance
At this plant, of course, installed a bunch of sensors that report accidents. However, a second confirmation loop was needed, because before the start of the project, a person necessarily resorted to the shop or point and visually confirmed the situation. Before that, it was impossible to take any critical measures.
')
This is a color frame from one of the cameras. Everything, in general, everything is covered with gray dust.
In contrast to the warm and attractive offices and sterile data centers, on this site constantly sharashit electromagnetic radiation from cables with a leg in thickness, interference from start-ups of different equipment. And then there is a great dust that accumulates centimeters per day.
Formulation of the problem
So, there is a set of sensors at the factory that signals some abnormal situations to the dispatcher. Since the sensors very often behave strangely, plus they give a far from complete picture of the scale of the situation and the nature of its development in general, in order to take a serious decision, you need to send a person to see what is happening. An intercom dispatcher calls the supervisor, and he is already looking for a brigadier and asking to be sorted out. There are two answers: “No, the computer drives” and, less often, “Mikhalych, it's ####, let's go #### urgent!”. A few minutes later comes a verbal description of the accident. If the accident is serious - in a very concise, capacious and figurative form. It is this form (delayed by 4-5 minutes) that causes the stop of the conveyor, emergency pressure relief on a number of units, breakage of expensive equipment during an emergency shutdown, and other things, after which the company usually suffers financial losses.
With financial losses, of course, it was impossible to put up.
Initially, the customer turned to us to calculate the cost of the machine vision system, which allows to automatically monitor all problem points. A list of nodes and a list of possible accidents was attached. For example, in this vat nothing should overflow; this node should not work as long as people are nearby; out of curiosity, bears sometimes come here to the pipeline; here there is a gap in the pipeline, and here it is necessary to ensure that a large piece of garbage, garbage — or tape does not break — does not go along the conveyor.
Looking ahead a bit, I'll show you a frame from the line where, for example, you need to control passing crowbars, sticks and another large-sized one. In the west, they still control the size of the fraction, but this is already much more expensive because of the requirements for the picture, the lighting, and because of the features of the code:
For such systems, there are ready-made industrial solutions developed for American, German and Japanese productions. Their main problem is that each non-standard detector requires a serious amount of code and tests for recognition. We talked with both domestic and Western developers of similar systems, and considered the first budget. The main problems are these non-standard detectors and plots that are difficult to recognize as above with almost no contrast.
The budget was very large, so the customer left to think further. According to his calculations, it turned out that it is much cheaper to raise a control tower in the 10-year perspective and put people who know the production like the back of their hand and are perfectly able to recognize abnormal situations. Plus, tighten the basic automation like the focus on the picture for triggering the sensor, basic motion tracking in areas where there should be no movement, etc. In general, a person should be - like a grandmother on an escalator in the subway.
So, we assembled the system on the basis of cameras, simple sensors and, in the program part, basic (library) detectors.
Technological vision
Correctly the system is called "technological vision". The technologist remotely looks at what is happening there at the plant and does not enter the aggressive environment. But in this aggressive environment of the plant are our cameras. The concept of technological vision still includes such solutions as specialized analytics on a node. For example, there is such a machine, the task is to mix the components and get a certain type of foam at the exit. Machine vision determines the size of the cell foam - depending on the size of the cell, the analyst determines which reagent to add and immediately gives a command to the machine controller. We have not yet implemented this, but, in general, it is quite real. Another thing is that, I repeat, it is here that, due to the special conditions, a living technologist is more economical.
The first problem of the project was that it was necessary to make a specific project proposal. If on a regular project it can be done in a week of sitting with plans and models and one departure, then it did not work out this way. We went there several times. We walked around each shop, talked with technologists, asked everyone what they needed to see and how they wanted the sensors to be located.
Here is a picture on the phone with a survey and a delineated field of view of the camera
The dimensions of the plant are huge (tens of kilometers), there are not always precise layouts, there is no possibility to attach to something on site to make a plan for installers in the future. As a result, we simply removed certain places and marked points for the cameras and another weak point on them. Like that:
The point is to remove not only the monazh point, but immediately together with the technologist what the camera will see. To do without neponyatki.
We did this for about two weeks. Typically, the design goes so much, but it took 2 months to complete the project. I must say that I walked around the facility with a service notebook, which during this time fell a couple of times, crashed, and even six months after putting the system on it, there were still traces of local dust.
Switching itself was relatively simple. The complex contains a ring of optics with branches, so there are no problems with the channel for the HD picture. Optics are there, by the way, because copper simply will not work. Long crossings between workshops are possible only on existing cable racks. And there are power cables with a cross section of the core in several centimeters. And there are not a dozen of them living there. Copper was only from shields to the cameras themselves, and in the cable with several levels of shielding.
The shields were assembled by themselves - they calculated the temperature regimes and put industrial noise-proof switches inside. As practice has shown, it works great, and in no way becomes a bottleneck. The problem was that there are no cameras as such for an industrial very hard use. In the sense that they still have the appropriate certificates. Therefore, again, it was necessary to take in special housings, to make grounding, to fasten special "wipers" from above to shake off the dust. Housings with wipers were found in one large vendor very well and immediately got up as it should.
Of the 165 cameras on commissioning, 19 fell off immediately. This transportation or factory defect is unknown. It was necessary to check in Moscow after arrival, but the customer needed as quickly as possible, so we sent immediately to the complex without unpacking.
When mounted their tracks, we learned what legacy-isolation is. There are many lines at the plant that have not been used since the times of the USSR, because they are duplicated, replaced and not unmounted, etc. In most cases, the customer gave us trays, and we mounted them. But on a number of runs there were consoles with large distances between the supports. It felt great thick self-supporting cable, and everything else was attached, in fact, on it. So, after 20 years of operation in an aggressive environment, the insulation on such cables turns into dust. While standing and watching - everything is fine. You touch the braid - it is poured with a quiet rustle.
In one of the legacy workshops
Total
Previously, a pipeline breakthrough could erode an embankment at a nearby lake in 15–20 minutes. Now the blur time has not changed, but the reaction time has changed. Visual confirmation can be obtained immediately and immediately make a decision. Previously, this required to jump into an all-terrain vehicle and go to the other end of the territory, for example. There is still need to add that it is not scary on this lake from the complex to walk only during both weeks of summer. All the rest of the time there are hot parties for addicts addicts. These animals adore strong unusual smells - they are like bears in Habr's reading at IT people. Therefore, checking the sensor every time was somehow not very interesting. We were also told stories about wolves, but we did not see them.
In addition to reducing the reaction time, the length of downtime after the accident also decreased. Previously, a big problem after a stop was to check that all repairmen left the site. This was done along the same chain and with difficulty. Now there is still a visual confirmation.
References:
- Video surveillance systems for data centers: who touched the rack and why
- Number Recognition (2012)
- Face recognition for access control systems
- Mail, if suddenly there are questions not for comments - DNikiforov@croc.ru
Source: https://habr.com/ru/post/265523/
All Articles