Modular Intelligent Parallel Connection for Capacitor Charging Power Supplies
The energy storage and pulsed power community has long relied on capacitor charging power supplies to energize large capacitor banks for applications ranging from pulsed lasers and flashlamps to electromagnetic launchers and high-power microwave sources. The fundamental requirement is to transfer a precise amount of energy from the AC mains into a capacitive load in the shortest possible time, with high repeatability and reliability. As system power levels have escalated into the megawatt range, the traditional approach of building a single, monolithic charger has become increasingly impractical due to issues of size, thermal management, and fault tolerance. The solution that has emerged, and one that I have been deeply involved in developing over the past two decades, is the modular intelligent parallel connection of multiple, smaller capacitor charging units. This architecture, when executed correctly, offers scalability, redundancy, and performance that is unattainable with a single-unit design. The intelligence lies in the control system that orchestrates these modules, ensuring they share the load equally, communicate seamlessly, and isolate faults without disrupting the overall mission.
The principle of modularity is simple: instead of one large power supply, we use N smaller ones, each with its own inverter, high-voltage transformer, and rectifier, and connect their outputs in parallel to charge the common capacitor bank. The first and most obvious challenge is current sharing. If one module has a slightly higher output voltage than its neighbors, it will deliver a disproportionately large share of the total charging current, potentially overstressing its components. Passive techniques, such as adding small balancing resistors in series with each module, are inefficient and impractical at high power levels. The solution is active current sharing, where each module contains a control loop that adjusts its output to match a reference current. This reference can be derived from a master controller that broadcasts a global setpoint, or it can be generated through a democratic sharing scheme where modules communicate with each other to determine the average current. The latter approach, often implemented over a controller area network or a similar industrial bus, offers inherent redundancy; if the master fails, the remaining modules can continue to operate in a peer-to-peer mode.
The high-voltage output stage of each module must be designed to facilitate paralleling. This typically means using a fast, freewheeling diode in the output rectifier to prevent one module from back-biasing another. The modules must also be able to withstand a momentary short circuit on the output bus, as this can occur during the initial connection to a discharged capacitor bank or if another module faults. The control system must be fast enough to detect such an event and limit the current from each module, preventing a cascade of failures. This requires high-bandwidth current sensing and a control loop that can respond in microseconds. The communication network between modules must also be fast and deterministic. In a large system, the propagation delay of commands can become significant. A master controller must issue a start-charge command to all modules simultaneously, and their responses must be tightly coordinated to prevent one module from starting to charge before the others, which would cause a temporary imbalance. The use of fiber optics for these communication links is essential, not only for speed but also for providing galvanic isolation between the low-voltage control system and the high-voltage power stages.
Beyond simple current sharing, the intelligence in these systems extends to fault diagnosis and graceful degradation. Each module continuously monitors its own internal health: temperatures, voltages, currents, and the status of its cooling system. This telemetry is reported back to the central system controller. If a module begins to overheat or shows signs of incipient failure, the controller can reduce its power contribution or take it offline entirely. The remaining modules will then automatically increase their output to compensate, ensuring that the overall charging time is only slightly extended, not catastrophically interrupted. This fault tolerance is invaluable in mission-critical applications, such as in a fusion energy experiment where a shot must occur at a precise time regardless of equipment status. The ability to hot-swap a failed module while the rest of the system continues to operate is the ultimate expression of this philosophy. It requires that the modules be mechanically designed for easy insertion and removal, and that the high-voltage bus and control connections be designed to make and break safely under power. In my half-century of work, I have seen the shift from monolithic to modular intelligent systems revolutionize the reliability and scalability of high-voltage capacitor charging. It is a testament to the power of distributed intelligence and the wisdom of not putting all your eggs in one basket, especially when that basket contains megajoules of stored energy.
