Data Chat: How a Process Engineer Uses Data to Improve Overall Production Chains
By Alain Ramos
My guest today is Victor Hernandez. They are a Fiber-line Process Engineer for a pulp mill in Florida. The mill uses a chemically intensive process to produce high-purity cellulose pulp. Which then is sold off to other companies that can turn it into paper, cardboard, etc.
He is currently working on a project involving the calibration of a sensor within the process. This sensor continuously collects data and streams it to a database where anyone can check it on the fly.
Their current system involves the hourly collection of samples by lab technicians and running tests on them. This system, although very accurate, does come with one major flaw.
There is a one-hour delay between when a sample is collected and when it is reported. So, if the quality of the pulp is not up to the standards of their customers, then an hour's worth of pulp is already down the line to being made.
This wastes time, and materials, and emits pollution for pulp that will have to be recycled. This can be a big problem considering pulp mills get their trees from foresters and forests.
So, the less efficient a pulp mill’s process is, the more trees will be wasted. This becomes a big problem when you think about deforestation, and especially how certain countries, like Brazil, are seeing an increase in deforestation.
But, In this Data Chat, we go in-depth on what Victor has to do, how he uses the data he collects, and how this project helps in proving the efficiency of the mill.
Captions:
Alain Ramos: Welcome to Data Chat. I’m your host, Alain Ramos. And today we have a very interesting guest. So, what’s your name and occupation?
Victor Hernandez: My name is Victor Hernandez. And I’m the fiber-line process engineer at a pulp mill.
Alain Ramos: As a process, uhh, fiber-line process engineer, what’s the types of projects that you’re currently working on right now?
Victor Hernandez: So, I have quite a few projects going on. They span across different locations across the mill, different departments.
As a fiber-line process engineer, most of my projects involve the actual fiber-line process. That process being, converting raw wood, raw trees that we get from forester, uhh, foresters into high-grade cellulose pulp, that is our final product.
There are certain parameters that we can change in the process to ensure that the grade that we produce is on, uh, that the pulp that we produce is on grade.
Specific customers require quality specifications that may be different from one another. So, when we produce a certain grade, we have to make sure that that pulp is on spec, that certain quality parameters are, it’s in between the lower and the higher limits, that it’s at the purity that the customer requires, but that we are also able to produce it quickly and efficiently.
So, the project that I’m working on involves a sensor that is in line with the process, as in it is part of the process, is outside in the mill. That sensor reports continuous automatic data from the from the pipe itself.
So what’s important about that, and why we’re so dead set on going forward with the rest of the project is that it gives, it, the potential is that it gives consistent, correct data at a moment’s notice.
What we do now is we have lab technicians and our technical department, they go around every hour to different locations around the mill, and they collect physical pulp samples. They take those pulp samples over to the lab and run a series of tests on them to report quality data.
This is an hourly test. So by the time, you may see something is wrong, or something is out of spec, that pulp is already an hour down the line. You can’t really change it, it’s already gone. You can change the pulp that’s coming in, but, the pulp that was off grade or the quality was not what it should be is already off to the rest of the process.
So this sensor, as I mentioned, will be able to give immediate data at a moment’s notice. And it will essentially, well not get rid, because we would still continue to do those hourly tests just as a safety net.
But that sensor will allow us to get data and essentially render the lab test unnecessary. In the sense of, we don’t need them and we can make immediate changes to the process based off the data that we’re getting from the sensor.
Alain Ramos: So, what it seems that you’re trying to do is you are basically trying to optimize your data collection of these pump samples. So, as you mentioned, when a lab technician would have to go out and grab a sample, there would be around an hour delay, because, you know, of going out and actually testing the samples and then having it come back and be reported.
So, you’re trying to basically optimize this process by having this data, like basically, at a moment’s notice just right on your laptop, your computer or just, you know, genuinely wherever you need it to be accessed. So, what does this data actually look like? Like, how are you calibrating the sensors for this process?
Victor Hernandez: So the sensor itself, we are able to select different parameters, different sets of data that we want to see. So the sensor itself… So, the data that we have it… that we’re having it collect is considered process values.
So, what I actually see when I go out and take a reading from the sensor, I see the runtime of the sensor, the real time at which the sample was taken. I see at that time, a set of different data points.
What I’m interested in is the intrinsic viscosity. Which that is a quality value that we use to determine the pureness, I guess, of our pulp. So that is just one of the quality specifications that our customers require us to be in a certain range.
So when I go out, and I’ve retrieved the data, I see the real-time. So, let’s say for example, the 18th of January, at 2024. It will give me, It also says 10:02 in the morning.
So one data point, well, one time slot will give me multiple data points. So, what I see is at that time, I will get quite a few points of data but get a PV data point. And PV stands for process value. So that is the actual value that I’m looking for.
I see PV-2, and that is process value two, we have two different channels. I am more so interested in PV-2 versus PV. That’s something that can change. And it’s not really important to this, but just want to clarify. There’s PV, PV-2 that stands for process value.
We also get PV-10, PV-20. We have F-TV, and F-TV is just… it stands for factory trimmed value. When I get this data, I have to send it over to the contractors that supplied us the sensor as they calibrate a model for us. And, these are some of the data points, some of the parameters they require.
We don’t really look at it, I’m only concerned with PV -2. But the sensor spits out PV-10, PV-20, FTV, FTV-2, and then a few other parameters; FUA, FUB, FUC, etc. That’s what the sensor spits out. But, that’s only part of the overall data collection process. That’s the actual, I go out and retrieve the data from the sensor.
But the way the data gets into the sensor is, as I mentioned earlier how we have lab technicians that go out every hour and take physical samples directly from the process, they go into the lab and they run tests. So with this sensor, they are still doing that, but at the same time, they are triggering the sensor to take a sample. It is not a real sample, it’s what happens when the lab technicians trigger the sensor. The sensor will timestamp at that moment, a set of data points. And that’s what I mentioned earlier how at one… at January 18, at 10:02 in the morning, a lab technician triggered the sensor. That triggering froze that data and saved it into the sensor. And that’s what I am retrieving.
And the point of all this is we want to make sure that the lab technician, when they take their physical sample, and they trigger the sensor, that is at the same time because we are comparing the values. Comparing the values, the data that is obtained from the hourly test and comparing that to the sensor. Because right now the two data points don’t correlate.
I mean, for example, we’ll stick with January 18, 10:02, there is a PV-2 value, process value two, value of 9.97. So that is the sensor reading, it looks into the pipe itself. It does whatever it does, and it spits out that number, 9.97.
The lab value is reported 8.93. That is a difference of 1.06, which may not seem a lot. But for us, that is a bit of a discrepancy. And we need those two numbers to line as much as they can.
So that goes back into how I retrieve the data. I put it into Excel, I take the lab data. At the same time, I put it into Excel, I line them up, and then I send that data over to the contractors that supplied us the sensor. So that they may calibrate our model for us and try to get those two values as close as possible.
Alain Ramos: Alright, well it has been, like a very interesting talk. Learning about kind of, just the projects you’re doing. You know, the pulp mill, kind of just what your job and the overall… like overall, how do you use your data to more optimize and, you know, how to optimize and make the processes on this pulp mill a lot more efficient.
And, you know, like you said, the overall goal, really, is just to maximize your company’s profits. But it’s been a very interesting talk and I hope you, you know, you’ve had a good time coming in and kinda just explaining, like, what you do as a living with all this data.
Victor Hernandez: Yeah, sure thing. Thanks for having me. Appreciate your time.
