Wow! This project of yours brings back memories from the 1960s when I was then a young electronics research technician assigned to a dynamic fatigue testing machine project. That project used an hydraulic piston connected to a vertical load train and controlled by a high-performance servo valve. It had a static load capability of plus or minus (compression and tension) 500,000 Lbf and a dynamic range of ±100,000 Lbf at a 60 Hz sinusoidal rate for fatigue testing. It used a commercial load cell inserted in series with the load train to measure both tension and compression in large test specimens.
This machine was located in a large test facility at Wright-Patterson AFB and was to be used to perform dynamic stress and fatigue tests on new alloys being developed for use by the Air Force. Problem was: it didn't work. The California company that built it under a competitive low-bid, fixed-cost plus fixed-fee contract eventually went bankrupt trying to get it to perform to specifications. This left the Air Force with a huge "white elephant" in a laboratory filled with German materials-testing machines seized after World War II and sent to various places in America. Sometimes these machines came with a German-speaking operator, but most often not. This American-built machine was supposed to be an upgrade to the German machines, which after all were getting a little "long in the tooth" and virtually impossible to obtain replacement parts.
So, enter the University of Dayton Research Institute (UDRI) to the rescue! My first task was to get the load cells working within a very noisy electrical environment. A load cell consists of four low-resistance strain gauges, securely cemented on the web of an I-beam, and connected in a Wheatstone Bridge to four (or six) wires leading to the outside world. There are often temperature-sensitive resistors internally wired in with the Wheatstone Bridge to perform temperature compensation and minimize "drift" in the load cell output, but this was none of my concern as long as the load cell worked properly. There was no way to "repair" a defective load cell. The load cells we were using had a circular Amphenol-style screw-on connector mounted to their cylindrical case, which was about a foot long and nine inches or so in diameter. Threaded holes in each end of the internally-cased I-beam allowed the load cell to be assembled in series with the load train, above the test sample, which in turn was secured to the hydraulic piston.
Bear in mind that 1960s servomechanisms were all-analog. Digital systems were on the horizon, but it was still "pie in the sky" back then. This machine had both force-feedback as well as position-feedback for the hydraulic piston, although not both at the same time. Force feedback came from the load cell of course, and position feedback was accomplished with a very long linear variable differential transformer (LVDT) connected between the machine frame and the bottom half of the double-acting piston. There were also data strip-chart recorders and various purpose-built amplifiers and power supplies, most of which I eventually discarded in favor of commercial off-the-shelf (COTS) equipment. My goal was to amplify the small (millivolt) differential signals from the load cell to approximately ±10 volts for recording on the paper strip-chart recorder, and also for use as a negative feedback signal to control the tension or compression produced by the hydraulic piston on the vertical load train.
Fast-forward now to the 21st century. The electronics community has fully embraced digital technology, but there are still applications where analog designs shine. This is not one of them, but you do have to start with basic analog design, beginning at the load cell. Your project is complicated by the fact that you have chosen to use four load cells, each rated for 2000 tonnes, to weigh your silo biomass of as much as 6000 tonnes. The use of multiple load cells is quite common for measuring large loads because you can just sum the weight measured by individual cells. There are basically only two ways to do this: sum the differential analog signals from your four load cells, or measure the differential analog signals digitally and numerically sum the results. This is where the HX711 and your Arduino shine. However, you will need one HX711 or each load cell.
This amazing little patented chip would have been a Godsend in the 1960s and 1970s. It features a selectable-gain differential amplifier to amplify the milli-volt signal of the load cell to a sufficient level for conversion by a 24-bit analog-to-digital converter to a serial digital output. The Arduino can be easily programmed to sum these digital outputs from four load cells. The HX711 also has an on-chip voltage regulator, requiring just one external pass transistor, to provide a precision and stable 5.0 volt excitation voltage for the load-cell bridge.
I would not worry about the 9 volt rating of your existing load cells. This is most likely the maximum voltage that can be safely applied, not the actual voltage you should apply. There will be some decrease in sensitivity with 5.0 volts applied instead of 9.0 volts, bu this is easily "calibrated out" in software on the Arduino.