Data Center Simulation Controls Engineer

Join our team as a Simulation & Controls Engineer to develop and operate digital twins for data center liquid-cooling systems. You'll work with hardware-in-the-loop simulations to optimize data center operations and improve efficiency. Collaborate with cross-functional teams to integrate simulation technologies into our data center operations. This role requires expertise in data center operations and simulation technologies. About us Brick builds physics-based simulation and control software for liquid-cooled AI data centres. Our existing stack includes dynamic thermal models of production liquid-cooling plants (Modelica, exported to FMI and driven from Python), reinforcement-learning supervisory controllers for coolant distribution unit setpoints, and a statistical validation harness that scores a controller across hundreds of randomised plant instances against a no-AI baseline. We are now building the layer above that: a real-time, hydraulically-solved model of a complete liquid-cooling plant that closes the loop with real PLC control code — so that cooling sequences of operation can be tested, and can fail safely, long before a building is energised. We are hiring the engineer who will own that. Why this role exists AI data centres are moving to direct-to-chip liquid cooling faster than the controls discipline around them has matured. Today, cooling control sequences get proven during commissioning — on a live building, with tenant hardware in the racks, on the critical path. Nobody wants that. The remedy already exists as an established practice in automotive and in power systems: virtual commissioning and controller-hardware-in-the-loop. It is almost entirely unpractised for liquid-cooling plants. That gap is what we are building into, and there is real customer pull behind it. You would be joining a small team, early, with an unusual amount of ownership. What you will build 1. A thermo-fluid plant model with a genuine pressure/flow solve — not a lumped thermal network: • pump head–flow curves and affinity laws; variable-speed-drive ramp dynamics; staging, lead/lag and redundancy • pipe and fitting resistance; control-valve sizing to IEC 534 / ISA S.75 (Cv / Kv / Av); valve authority; actuator stroke times • check valves, bypass and decoupler arrangements, differential-pressure control, loop balancing • ε-NTU or effectiveness-map heat exchangers with fouling; CDU primary/secondary decoupling • closed-circuit and adiabatic heat rejection, including changeover logic and spray staging • thermal storage with stratification and charge/discharge rate limits • water/glycol mixtures with temperature-dependent properties — yes, viscosity matters • rack-side heat load driven by real AI workload power traces, including fast transients 2. Real-time execution. FMI-compliant export, fixed-step, deterministic, with a bounded solver budget per step. The model has to hold a hard deadline every step, not on average. 3. The controls interface. Bind the plant model to control code over industrial protocols — Modbus TCP, BACnet/IP, OPC UA, EtherNet/IP — starting with software-in-the-loop against virtual controllers (S7-PLCSIM Advanced, Logix Emulate, CODESYS), progressing to controller-hardware-in-the-loop against real PLCs, and ultimately to wired analog I/O (4–20 mA, 0–10 V, RTD). 4. A fault-injection library. Sensors stuck, drifting, out of range or lost; valves stuck mid-stroke; pump trips; VFD faults; heat-exchanger fouling; strainer blockage; flow maldistribution across parallel branches; leaks; loss of facility water; dropped, delayed and stale communications. 5. Test orchestration and evidence. Scripted, deterministic scenarios with pass/fail predicates tied to specific clauses of a sequence of operations; regression on every controller build; time-stamped output in a form a commissioning authority will accept. 6. Model validation you can defend. Calibration against measured plant data to a recognised statistical bar (ASHRAE Guideline 14-class CVRMSE / NMBE). A control decision validated inside a model is only as good as that model’s credibility, and you will be the person who has to stand behind it in front of a customer’s controls lead. What you must bring • Deep, hands-on 1-D thermo-fluid and hydraulic network system modelling. Modelica (Dymola, Modelon Impact, OpenModelica, the LBNL Buildings Library) is our default; equivalent depth in Simcenter Amesim or Flomaster, GT-SUITE, Apros or similar is welcome. You must have personally built and debugged a network model in which flow and pressure are solved, not prescribed. • FMI/FMU in anger — export, co-simulation versus model exchange, solver selection, and first-hand knowledge of what breaks when you try to make one run in real time: stiffness, algebraic loops, index reduction, tearing, event handling. • Industrial controls literacy. You have worked with PLCs, can read structured text or ladder, understand a scan cycle, and have moved real data over at least one of Modbus, BACnet, OPC UA or EtherNet/IP. • HVAC or process-plant grounding — chillers, pumps, heat exchangers, cooling towers or fluid coolers, control valves — at a level where you can read a P&ID and argue about it. • Python, and the ability to write software other engineers can maintain: version control, tests, CI. • The temperament for greenfield work with a customer clock running on it — and the honesty to say out loud when a model is not good enough to base a decision on. Strongly desirable • Hands-on virtual commissioning or hardware-in-the-loop delivery, in any industry. If you have built a rig, we want to hear about the plumbing, not the marketing. • Data-centre cooling specifics: CDUs, technology cooling system versus facility water system loops, ASHRAE TC9.9 liquid-cooling classes, Open Compute Project cold-plate and CDU work, direct-to-chip at 100 kW+ per rack. • Commissioning-process literacy: ASHRAE Guideline 0 / 1.1 / 36, Level 1–5 commissioning, functional performance testing, integrated systems testing, and a realistic sense of what a commissioning authority will and will not accept as evidence. • Real-time simulation targets — dSPACE, OPAL-RT, RTDS, Speedgoat, Typhoon HIL — or experience working with national-laboratory HiL facilities. • System identification and model calibration against measured plant data. • Exposure to microgrids, battery storage or demand-flexibility control. The cooling plants we model have to be dispatchable. What this role is not • Not a CFD role. CFD has a place upstream, in component design and in generating data for surrogates. The deliverable here is 1-D system simulation that runs faster than the plant it represents. • Not a machine-learning role. We already have controllers. What we are missing is the plant they get tested against. • Not maintaining someone else’s simulator. You will be choosing the architecture, and defending the choice. Your first six months • 30 days — you own the modelling architecture decision: toolchain, library, solver, real-time strategy, with a written justification we can put in front of a customer’s control systems lead. • 90 days — a single-loop plant model with a working pressure/flow solve, exported as an FMU, closing the loop with a virtual PLC running a real sequence of operations, plus a first fault-injection set and an honestly measured real-time factor. • 180 days — that rig running against real controller hardware, with a scenario library, pass/fail reporting and a validation write-up. How to apply Submit APPLICATION with your CV or profile and short answers to three questions: 1. Describe the most complex fluid-network model you have personally built. What was solved, what was prescribed, and what did you get wrong first? 2. How would you approach getting a Modelica hydraulic network to hold a fixed 5 ms step with a hard deadline — and how would you know when it simply cannot? 3. What have you shipped that talked to a PLC? No cover letter needed. We read every application.