USAID posted an article on their blog about 3D printing and the IEPAS weather station project.
We have been a little remiss in reporting progress on the MMA Weather Station over the last few months, but we’ve had some early prototypes out in the elements for testing and we learned some important lessons on durability and reliability. Here’s an update on what we’ve been up to:
Outer Shell and Structure
Aside from electronic components, parts for the stations are made on a 3D printer. This includes some of the circuit ‘boards’ and wire connectors. In late spring / early summer, we placed a station at the Marshall Field Site in Boulder, CO. The site is used by NCAR/UCAR to test stations and sensors. While it’s only been out at Marshall for few months, we were nonetheless pleased to see that the components show no sign of aging in the sun, wind, and rain. Connections between modules hadn’t moved, the plastic did not discolor or crack, and the movement of the anemometer remains unchanged. Since placing the station at Marshall, we’ve made a number of changes to further strengthen and improve the ruggedness of the design.
We struggled for some time with how ‘smart’ each station should be, as well as how best to perform basic data logging, data processing, time stamping and synchronization, and communications functions you would expect or need. We initially threw out any notion of developing our own circuits, as this would undermine the spirit of the project in several ways. First, we did not want to mass produce boards, as this then created one more highly specialized external dependency for the meteorological services that would be making the stations. Our goal is to enable the manufacture and assembly of weather stations, by using readily available off the shelf base components. This allows for long term maintenance, as well as customization. Secondly, if we opted for the boards and circuits to be built in-house by each NMHS, then this would require special training, skills, and equipment. If too much skill is required, then the ability for the system to work in areas of limited resources is threatened.
We therefore had the option of using some of the common hobby micro-controllers out there, such as the Arduino. While inexpensive and easy to setup, we eventually rejected such devices, due to the cost of add on components and features. For instance to add data logging would cost $15, wireless communications $30; so on and so forth.
Eventually we looked at single board computers, and chose the Raspberry Pi. It quite literally is a small computer. It comes with ports for Ethernet and USB connections, wireless functionality is added easily and cheaply, data can be logged to the same card that holds the operating system, and it still hosts all the GPIO functionality we need. As it is a full on operating system, we expect to be able to provide some more advance features, which make sense and are nice for remote use and access locations. Data processing can be done at the weather station, rather than waiting for collection and centralization of information. A GUI management interface is easy to create and provide. Updates of scripts (Python and shell) used to read the sensors, manage data, and create communication links are easily updated on the fly, as opposed to a micro-controller.
Currently, we have all the basic scripts and functionality for the station working on the Raspberry Pi. There is a lot of polishing and further testing to do, but at this point, we’re very pleased with the selection.
Junction Box and Circuits
The Junction Box gathers together all the cables from the various sensors and also houses the rain gauge amplifier, a real time clock (RTC) IC, and an analog to digital converter (ADC) for the analog sensors (temperature, humidity, rain and wind direction, necessary because the Raspberry Pi will only accept digital input). The introduction of the ADC, like the amplifier — a 16 pin IC chip, prompted a rethink of the “Steinson Circuit” that we were using with the amplifier. While the circuit seemed to be pretty reliable, it was somewhat fiddly to make and certainly had the potential to cause problems, so the prospect of having a “Steinson Circuit 2” for the ADC did not seem like a good idea. We still wanted to avoid soldered connections and preferred the use of off-the-shelf, push fit jumper cables to standardize the design of all the other multi-cable connectors that will link the various components of the weather station (see below). We also wanted to enclose and seal all the connections to avoid corrosion and temperature differential issues, which can be a problem with these types of sensor in this kind of application. We believe the new approach achieves all these goals while at the same time being easier to assemble and very robust (all connections are tightly clamped and enclosed under a sealing layer of hot glue).
Connectors and Plugs
We started off using CAT 5e punch-down jacks to connect different components of the weather station together but during initial testing it was clear that this approach would not be durable enough. We’ve developed a series of connectors which clamp together multiple push-fit jumpers in a way that allows easy disconnect of individual components while also providing very stable connections, thanks to the precision of the 3D printer and liberal application of hot glue to secure and seal. These connectors have evolved into plugs that are integrated into the cases of the junction box, rain gauge and radiation shield, providing a plug-and-play approach that should make assembly in the field foolproof and minimize disruption during maintenance (a replacement rain gauge could be swapped out in a matter of minutes without the need to open up either the new gauge or the junction box). The plugs are designed so that they can rotate in their socket, which allows a convenient way of leveling the component pieces before they are clamped in place (particularly important for the rain gauge).
Little of consequence has changed on the rain gauge since we originally described it (see Hybrid Tipping Bucket Rain Gauge Design), other than the introduction of the connector plugs described above. The plug in approach provides much better protection to the very delicate wires on the load cell, which tended to break easily in previous iterations.
Low Cost Anemometer
No self-respecting weather station can do without a means of measuring wind speed and direction. We have some ideas for a combined gauge that will do both at once but it will need some extensive development and testing, so in the meantime, we plan to include a conventional, three cup anemometer and wind vane in our station. We’re using a digital (latch) Hall Effect sensor triggered by four small magnets to count rotor revolutions for the anemometer (these will then be converted to wind speed by a script in the Raspberry Pi), and an analog Hall Effects rotational sensor, again triggered by magnet, to determine angular displacement of the vane. Once again, the majority of components are made on the 3D printer, and a small, sealed ball bearing allows low friction rotation.
We did some initial testing of the anemometer in our own “high tech testing center” (A.K.A. the warehouse with various fans and blowers.) and the results from three prototype units demonstrated consistent enough performance for us to move to formal testing. We later gave one unit a brief run in a wind tunnel, where it survived a wind of around 115mph (low category 3 hurricane), although there were some indications that it was not going to last much longer. With the lessons learned from the brief wind tunnel experiment we did some redesign so that the various components are locked in place and we are now ready for a full calibration test which we hope will be conducted by the National Weather Service Field Support Center in Sterling, VA.
The radiation shield protects various sensors (temperature, humidity, pressure) from direct exposure to the sun, while allowing sufficient circulation so that the air being sampled is representative of outside conditions. We have loosely based our design on the Maximum Minimum Temperature Shield (MMTS) that is used by the US National Weather Service, although the diameter of the leaves was dictated by the size of the build plate on the 3D printer. We’re trying something that we haven’t seen in commonly used shields – we’ve added a mesh screen to all air inlet/outlet passages to dissuade wasps etc. from making their home inside the shield. We figure that the remote locations that we are targeting with the MMA project will likely mean sporadic maintenance for many of the stations, so colonization by creepy crawlies is a distinct possibility that we should try our best to prevent. In addition to field testing, we also hope to eventually test all the components in the wind tunnel to see how they withstand extreme winds. All the components of the Radiation Shield are made on the 3D printer.
Sensor Connections and Holders
The various electronic sensors that we’re using all tend to have very delicate connectors, so to ensure reliable and consistent connections we’ve conceded that this is the one place where we cannot avoid the soldering iron. Fortunately, these are all simple, easy access soldering jobs that even the dexterity-challenged (such as your humble scribe!) can handle.
After several iterations, we’ve narrowed down the hybrid gauge to two versions that we are going to subject to initial testing. The first version, modestly named the “Ultimate” is the largest version of the gauge that we can make in the confines of the 5” x 5” x 5” build volume that our current, preferred 3D printer allows. Other 3D printers have a larger build volume, but the model of printer we have selected we feel is more reliable for field projects. The second version of the hybrid gauge, prosaically named the “4’’ Pipe” is designed to fit inside a standard 4” PVC pipe, so that the pipe acts as both the casing and the funnel rim, two of the largest and most time consuming parts to manufacture. This means that the bucket used in the 4” Pipe is significantly smaller than in the Ultimate, and if you’ve had a chance to read our fascinating analysis of the hybrid gauge design (see Hybrid Tipping Bucket Rain Gauge Design) you’ll know that one of our theories is that a larger bucket results in less tips, which reduces one of the sources of error that afflicts tipping bucket rain gauges; i.e. the tendency for rain to enter the bucket after it has begun to tip. All other things being equal, we think this gives the Ultimate a slight edge over the 4” Pipe, although the smaller bucket does allow us to use less gain on our signal amplifier, which means that we can get better reading resolution (more on this to come in a later article). To determine conclusively which is better, we will need to conduct some extensive field tests, which we hope to do in the coming months. Right now we just want to make sure that our concept is sound so that we aren’t laughed out of the testing center when we eventually get there, so here goes…….
Before we could start testing, we had to settle on how some of the electronic circuitry would be configured, at least temporarily, so that we could receive and process the data coming out of the gauge. Up until now, we have been using an Arduino board to read the output from the load cell and, for the moment, we’re going to continue with it, although we eventually expect to migrate to a Raspberry Pi platform for added flexibility and computing power. There’s an article elsewhere on the site about the amplifier circuit that we are using for the load cell (see 3D Printed Circuit Board) and we decided to house this in a central junction box that will also serve as the mounting platform for the rain gauge and the other gauges/sensors that we eventually hope to deploy, which should include an anemometer and a “Stevenson Box” to house multiple sensors, such as temperature, humidity, pressure, etc. Also in the junction box are a couple of CAT 5e punch-down jacks that we will use to connect the various components together (for now, load cell to amplifier; amplifier to Arduino board – see Photo 1).
Until now, we hadn’t attempted to read more than a few seconds worth of data from the load cell, so storage for later evaluation hasn’t really been an issue, but for these tests to be meaningful, we needed to be able to collect several hours’ worth, which can really add up when you’re sampling as frequently as 10 times per second. To transfer the data from the Arduino to a laptop, we used PuTTY ( http://www.putty.org/) which is “a free and open-source terminal emulator, serial console and network file transfer application”. Some initial fiddling with the load cell and our first test indicated that the output was a little jumpy, so I was very happy to find a simple little smoothing routine in a tutorial from our friends at Arduino (http://arduino.cc/en/Tutorial/Smoothing).
All we needed now was a rig on which to mount the gauge and a supply of simulated rainfall. It turns out that even torrential rain doesn’t look like a lot of water when it’s condensed into a stream that represents the amount falling within the collection orifice of a typical rain gauge, so the first requirement was a means of controlling water flow to a very slow drip. Rather than using a 1000 words to describe how we did it, take a look at photo 2. The mount for all this was cobbled together from an old lamp stand and various PVC and GI pipe pieces and fittings (photo 3).
Our initial testing has focused on pouring a measured amount of water through the gauge (measured by weighing the contents of the water flask before and after the test) and then totting up how much water the gauge has recorded passing through. So far the results have been pretty encouraging – all within 2% using a pretty rudimentary approach for data analysis – so we’re going ahead with the planning for some real field testing.
IEPAS’ project partner and activity lead, the Consortium for Capacity Building, has released a draft a USAID OFDA funded study. The report, “Hydro-Meteorological Disaster Risk Reduction: A Survey of Lessons Learned for Resilient Adaptation to a Changing Climate“, examines a decade of work to find lessons for climate adaptation projects. In capacity development activities, what constitutes climate adaptation is often difficult to define. As a result many adaptation activities are ‘business as usual’ or repeat investments made earlier. The report from the CCB looks for best practices from hydro-meteorological disaster risk reduction projects to provide guidance for those investing in climate adaptation.
Local Copy / Link:
As described in more detail elsewhere on this site, the MMA project aims to take advantage of the recent emergence of low cost electronic components and innovations in micro manufacturing technologies to develop a range of affordable meteorological instruments for manufacture and use in developing countries.
The automatic rain gauge that is currently under development aims to improve the performance of the traditional tipping bucket gauge (the most commonly used automatic precipitation gauge) by adding the capability to weigh the rainwater as it is collected in the bucket using the components of inexpensive (but very precise) weighing scales that are now commonly available. Using this approach, the tipping bucket itself no longer plays a part in measuring rainfall but simply provides a means of emptying the collection vessel in a controlled way. By measuring the weight of water in the bucket at regular intervals and subtracting the previous reading, the volume of water collected per second or minute can be determined and translated into mm/inches of rainfall.
We’re calling it the hybrid gauge because it combines some of the features of the tipping bucket with some of the features of an automatic weighing gauge, which usually takes the form of a large container (that does not automatically empty) and an elaborate spring balance/drum recording device. As far as we know, this is a new idea – the following section lists some of the advantages that we see with this approach.
Advantages of the Hybrid Tipping Bucket Design
Section 220.127.116.11 of WMO’s CIMO Guide Part 1: Measurement of Meteorological Variables lists six main sources of error specific to tipping bucket rain gauges:
(a) The bucket takes a small but finite time to tip and, during the first half of its motion, additional rain may enter the compartment that already contains the calculated amount of rainfall;
(b) With the usual bucket design, the exposed water surface is large in relation to its volume, meaning that appreciable evaporation losses can occur, especially in hot regions. This error may be significant in light rain;
(c) The discontinuous nature of the record may not provide satisfactory data during light drizzle or very light rain. In particular, the time of onset and cessation of precipitation cannot be accurately determined;
(d) Water may adhere to both the walls and the lip of the bucket, resulting in rain residue in the bucket and additional weight to be overcome by the tipping action. Tests on waxed buckets produced a 4 per cent reduction in the volume required to tip the balance compared with non-waxed buckets. Volumetric calibration can change, without adjustment of the calibration screws, by variation of bucket wettability through surface oxidation or contamination by impurities and variations in surface tension;
(e) The stream of water falling from the funnel onto the exposed bucket may cause over-reading, depending on the size, shape and position of the nozzle;
(f) The instrument is particularly prone to bearing friction and to having an improperly balanced bucket because the gauge is not level.
We believe that the hybrid Tipping Bucket/Weighing Gauge that the IEPAS project is currently developing has the potential to improve on all of these 6 sources of error, when compared to a standard tipping bucket; here’s our logic:
(a) To be sensitive to light rainfall, the standard tipping bucket rain gauge tips at 0.1 – 0.2 mm of rain (per the CIMO guide, “This amount of rain should not exceed 0.2 mm if detailed records are required”). The problem with this is that during heavy rainfall, the bucket tips back and forth like a Ramones’ metronome – by our calculation, 250mm/hr rainfall would require 1250 tips per hour, which works out at one tip every 2.9 seconds. The hybrid gauge does not use the number of tips to measure the rate of rainfall; instead, it continuously weighs the rainwater accumulating in the bucket and uses the tipping action simply for emptying the bucket. This means that the gauge can be designed to tip far less regularly – the largest version of the hybrid gauge should (conservatively) tip after around 2mm of rain (every 29 seconds at 250mm/hr), so without attempting to address the cause of the error, the hybrid gauge reduces the frequency of occurrence by around 90%. The increase in rainfall required per tip is achieved by using a larger bucket and by reducing the size of the collection orifice. According to the CIMO Guide, “The size of the collector orifice is not critical for liquid precipitation, but an area of at least 200 cm2 is required if solid forms of precipitation are expected in significant quantity”, so as long as we avoid installations such as the top of Mount Kilimanjaro, the hybrid gauge with a smaller orifice (75cm2) should be fine for the tropical/sub-tropical locations for which it is intended.
(b) The hybrid gauge calculates the amount of rainfall by measuring the increased weight of water that accumulates in the bucket over a set period of time. Because the weighing scale is sensitive to one tenth of a gram, this time period can be very short (for 250mm/hr rate of rainfall, 0.1 grams will accumulate in around 0.2 seconds) so evaporation should not have a significant effect on readings, even in very light rainfall. (2.5mm/hr rate of rainfall should be detected by the hybrid gauge in 20 seconds, as opposed to almost 5 minutes for a standard gauge).
(c) The method of measurement and sensitivity of the hybrid gauge allows it to keep a continuous record even in very light rainfall, so onset and cessation of rain will register quite precisely in all but the very lightest and shortest events – the “no tip” situation that occurs when a small amount of rain falls that is not sufficient to trigger the standard gauge is not a problem with the hybrid.
(d) ,(e) and (f) These issues are related to the need for the standard gauge to tip with exactly the same volume of water every time, as the number of tips is used to calculate rainfall totals. The hybrid gauge measures the difference in water mass collected in the bucket from second to second, so it doesn’t really matter if the tipping action is inconsistent, as it is not used to calculate rainfall. The added advantage of this is that dimensional tolerances and precision in manufacturing, assembly and installation are less critical in the hybrid gauge, making it much more suited to local production.
Section 6.4 of the CIMO guide lists the major causes of error that are common to all types of precipitation gauges, namely:
(a) Error due to systematic wind field deformation above the gauge orifice: typically 2 to 10 per cent for rain and 10 to 50 per cent for snow;
(b) Error due to the wetting loss on the internal walls of the collector;
(c) Error due to the wetting loss in the container when it is emptied: typically 2 to 15 per cent in summer and 1 to 8 per cent in winter, for (b) and (c) together;
(d) Error due to evaporation from the container (most important in hot climates): 0 to 4 per cent;
(e) Error due to blowing and drifting snow;
(f) Error due to the in- and out-splashing of water: 1 to 2 per cent;
We believe that these potential errors relate to the hybrid gauge in the following way:
(a) and (f) are independent of the hybrid’s measurement approach and so are indeed common with standard gauges (we are attempting to minimize them in the design of the collecting funnel).
(b) By using a smaller orifice (and hence smaller funnel) than the standard gauge, the surface area of the internal walls of the collector are significantly reduced, which should reduce the wetting losses. This advantage is countered to some extent by the rough surface finish that results from the method of manufacture; it’s possible that we may use some treatment (e.g wax) on the wetted surfaces to improve this.
(c) As we are measuring the change in mass in the bucket from second to second, this is not relevant.
(d) As described above, this should be minimal for the hybrid gauge
(e) We don’t measure no stinkin’snow.
Sources of Error Unique to the Hybrid Gauge
Right now only three potential sources of error come to mind that are related to the use of load cells to weigh the accumulated rainfall.
(a) Inconsistent output/load response between different load cells
(b) Non-linear output from the load cell under very light load
(c) Floating “empty bucket” output
(a) Initial testing (and literature review) suggests that even the inexpensive load cells being used are extremely consistent in their response to load, to the extent that calibration of individual gauges is probably unnecessary
(b) The design of the hybrid gauge is such that the load cell is pre-loaded by the weight of the bucket and it’s support structure, so it’s always operating in its linear range
(c) It is possible that the “empty bucket” output might drift very slightly from day to day, which might give a false indication of a minute amount of rainfall (can probably be discounted during quality assurance of the data?)
Because our own Mr. Steinson would not claim the credit for himself, I give to you the ‘Steinson Circuit Board'; a circuit board made with a 3D printer using ABS plastic.
For size reference, a U.S. quarter is just slightly smaller than the square. To complete it a resistor, amplifier IC (placed on the opposite side of board), and connecting wires are inserted into slots and holes in the plastic board, making a perfectly working circuit. While it may look particularly shiny, no glues, conductive paints, or other materials were used to fasten the wires and connectors. The pins for the amplifier are simply bent, as are the leads for the resistor. Wires are ‘hammered’ in place with a small awl. Once put together, the wires and components are surprisingly secure. In some ways the board offers more protection than a typically soldered PCB. Not shown in the picture above is a small cover and clamp that prevents accidental shorts (such as by laying the board on a conductive surface) and further secures the wires and components.
It might seem that we used a new tool to do something that can be better accomplished with other approaches. For the MMA weather station project, however, we dreaded the idea of designing and ordering custom PCBs, as this would undermine the whole idea of bypassing mass-production in favor of local fabrication and eventual customization. A small CNC router could make PCBs, but small boards are difficult to build, and it added another tool and level of complexity to the project. Again, the point of the MMA is to move fabrication and eventual design of weather stations to developing countries. Similarly, the thought of etching a PCB with chemicals was unattractive given the skills necessary, likely scarcity of the materials needed in our envisioned project locations, and some danger / hazard to the maker. Of course solder boards are available, but then too we would need to supply boards externally, as well as introduce new skills and tools.
Most of the circuits on the weather station are very simple; typically consisting of a sensor, amplifier, resistor and then points for power and read out. Few of the sensors the project has considered are ICs, and for now at least there are no SMD components. The only chips in use are the amplifiers and clock timers. Everything then hooks up to a micro-processor, such as an Arduino. While the 3D printed plastic circuit is not for every application, the ability to do so makes the stations considerably more simple and inexpensive to fabricate. It also makes the overall project simpler, by not necessitating additional tools and methods to address one specific component.
Eh….probably not….but maybe.
Some of the structural components we have designed for the weather station have been developed on a 3D printer that uses PLA. PLA is great as the warping is less than with ABS. This is particularly important for larger pieces or those with extremely thin walls, such as the Stevenson screen. Typically corners are printed more sharply as well, and PLA is comparably more rigid than ABS. All of this is good for the end ‘fit and finish’ of the station, or case components for something like a Chatty Beetle.
Unfortunately, PLA is not really intended for outdoor use. One of the great environmentally friendly things about PLA is that it is biodegradable, but I am not so sure we will want this material property for a weather station. Some of the research looking at the properties of PLA note that biodegradation will occur in a matter of months, but this is with powder or small fibers maintained at a high temperature and moisture in a controlled composter. A larger piece undergoing natural temperature and humidity fluctuations, such as for the project weather station, might not biodegrade as quickly.
Despite many stating that PLA is not appropriate for outdoor use, which may be true for structurally critical components, the survivability may just be fine for light structural applications. The components on our weather station, for instance, are largely to hold sensors and wiring that is feather weight. In Biodegradable Poly (Lactic Acid): Synthesis, Modification, Processing, and Applications, author Jie Ren notes:
To biodegrade within 90 days, as described, the products have to reach 140 F for 10 consecutive days. This requires a special facility, which few consumers have access to. If your PLA products end up at the landfill, they will not degrade any faster than a petroleum-based product.
The above mentioned reference does, however, state that PLA will degrade in high humidity and temperatures above 110 F. In all likelihood then, PLA does degrade when outside, but the rapid biodegradation often discussed requires a specific set of conditions, which are unlikely to naturally occur. As a side note, PLA is referenced as considerably UV resistant.
So the question is will a weather station printed with PLA degrade considerably over a one year period, or will it be able to remain largely intact up to a three year period, when replacement of structural components makes sense in the normal life cycle of the station? Based upon some of the references I have mentioned, I do not think PLA would rapidly deteoriate as shown in a composting situation, but it is not clear if some level of degradation would still occur, which would make a component unusable or affect the readings taken by a sensor. Of course the durability will be affected by local environmental conditions, to which the PLA printed structures are exposed.
To attempt to figure this out, we are placing test blocks of ABS and PLA in various outside conditions, however, I suspect in the end, we will initially need to print in ABS, as we do not want to delay the project simply to test structural components over a multiple year period.
Injection molding provides a number of advantages over most additive fabrication techniques, such as 3D extrusion printers. Even with small, manually operated appliances, the parts produced on an injection mold machine will have a better finish and greater strength than those produced by a filament extrusion 3D printer. Also, injection mold machines can typically support a wider variety of materials, such as various rubbers.
Of course the real advantage of injection mold processes is speed. Once a mold is created, a part can be finished in minutes, if not seconds. Consider a button. These can be produced in seconds on even a manually operated injection mold machine. On a 3D printer, you should expect 15-30 minutes; sometimes longer. Printing six 1cm x 1cm x 1cm calibration cubes (at once) on our various printers takes anywhere from 2-3 hours.
Getting the speed advantage of an injection mold machine, however, requires an up front investment in designing, testing, and then finalizing a mold for the desired part. This is not trivial.
For our weather station project, we do not foresee any country placing more than a hundred stations, and we would claim victory after a couple dozen fabricated, calibrated, fielded, and operating stations. In our envisioned application, slow production of stations is okay, if not desirable. The intent is not to deploy 100 stations at once, but rather to do so over months and even perhaps a couple years. This has several advantages, but the most important benefit is not outstripping the absorptive capacity of a national weather service to place and maintain equipment. Another advantage is an ability to slightly iterate and customize parts, as units are deployed, thereby integrating lessons learned in the middle of a project.
Having weighed options, an injection mold machine in the MMA project would likely only be used for highly detailed parts, or those that require special material. Otherwise, the investment in a high quality mold outweighs the benefit.
RANET / IEPAS systems, such as the Chatty Beetle, RAPIDCast, and RAW SMS, will be unaffected by the Government shutdown. IEPAS operates under a Cooperative Agreement / grant, and no systems are on or tied to Federal facilities. Information feeds from Federal partners, which our systems rely upon, such as for tsunami, are likely to be considered essential, and so again IEPAS activities will remain unaffected.