Three-phase electrical power is a critical technology for heavy machinery. Learn how these US Coast Guard Academy students built a physical generator set model capable of producing three-phase electricity. The article steps through the power sensors, master controller and DC-DC conversion design choices they faced with this project.
Three-phase electrical power is typically used by heavy machinery due to its constant power transfer, and is used on board US Coast Guard cutters to power shipboard systems while at sea. In most applications, electrical power is generated by using a prime mover such as a diesel engine, steam turbine or water turbine to drive the shaft of a synchronous generator mechanically. The generator converts mechanical power to electrical power by using a field coil (electromagnet) on its spinning rotor to induce a changing current in its stationary stator coils. The flow of electrons in the stator coils is then distributed by conductors to energize various systems, such as lights, computers or pumps. If more electrical power is required by the facility, more mechanical power is needed to drive the generator, so more fuel, steam or water is fed to the prime mover. Together, the prime mover and the generator are referred to as a generator set “genset”.
Because the load expects a specific voltage and frequency for normal operation, the genset must regulate its output using a combination of its throttle setting and rotor field strength. When a real load, such as a light bulb, is switched on, it consumes more real power from the electrical distribution bus, and the load physically slows down the genset, reducing the output frequency and voltage. The shaft rotational speed determines the number of times per second the rotor’s magnetic field sweeps past the stator coils, and determines the frequency of the sinusoidal output. Increasing the throttle returns the frequency and voltage to their setpoints.
When a partially reactive load—for example, an induction motor—is switched on, it consumes real power, but also adds a complex component called “reactive power.” This causes a voltage change due to the way a generator produces the demanded phase offset between supplied voltage and current. An inductive load, common in industrial settings, causes the voltage output to sag, whereas a capacitive load causes the voltage to rise. Voltage induced in the stator is controlled by changing the strength of the rotor’s electromagnetic field that sweeps past the stator coils in accordance with Faraday’s Law of inductance. Increasing the voltage supply to the rotor’s electromagnet increases the magnetic field and brings the voltage back up to its setpoint.
The objective of our project was to build a physical generator set model capable of producing three-phase electricity, and maintain each “Y”-connected phase at an output voltage of 120 ±5 V RMS (AC) and frequency of 60 ±0.5 Hz. When the load on the system changes, provided the system is not pushed beyond its operating limits, the control system should be capable of returning the output to the acceptable voltage and frequency ranges within 3 seconds. When controlling multiple gensets paralleled in island operation, the distributed system should be able to meet the same voltage and frequency requirements, while simultaneously balancing the real and reactive power from all online gensets.
Gensets supply power in two conceptually different configurations: “island” operation with stand-alone or paralleled (electrically connected) gensets, or gensets paralleled to an “infinite” bus.” In island operation, the entire electrical bus is relatively small—either one genset or a small number of total gensets—so any changes made by one genset directly affects the voltage and frequency of the electrical bus. When paralleled to an infinite bus such as the power grid, the bus is too powerful for a single genset to change the voltage or frequency. Coast Guard cutters use gensets in island operation, so that is the focus of this article.
When in island operation, deciding how much to compensate for a voltage or frequency change is accomplished using either droop or isochronous (iso) control. Droop control uses a proportional response to reduce error between the genset output and the desired setpoint. For example, if the frequency of the output drops, then the throttle of the prime mover is opened correspondingly to generate more power and raise the frequency back up. Since a proportional response cannot ever achieve the setpoint when loaded (a certain amount of constant error is required to keep the throttle open), the output frequency tends to decrease linearly with an increase in power output. A no-load to full-load droop of 2.4 Hz is typical for a generator in the United States, but this can usually be adjusted by the user.
Frequency control typically uses a mechanical governor to provide the proportional throttle response to meet real power demand. Voltage control typically uses an automatic voltage regulator (AVR) to manipulate the field coil strength to meet reactive power demand. Isochronous mode is more challenging, because it always works to return the genset output to the setpoint. Maintaining zero error on the output usually requires some combination of a proportional response to compensate for load fluctuation quickly, and also a long-term fine-tuning compensation to ensure the steady-state output achieves the setpoint.
If two or more gensets are paralleled, the combined load is supplied by the combined power output of the gensets. As before, maintaining the expected operating voltage and frequency is the first priority, but with multiple gensets, careful changes to the throttle and field can also redistribute the real and reactive power to meet real and reactive power demand efficiently.
If the average throttle or field setting is increased, then the overall bus frequency or voltage, respectively, also increases. If the average throttle or field setting stays the same while two gensets adjust their settings in opposite directions, the frequency or voltage stay the same, but the genset that increased their throttle or field provides a greater portion of the real or reactive power. Redistribution is important because it allows gensets to produce real power at peak efficiency and share reactive power evenly, because excessive reactive power generation derates the generator. Reactive currents flowing through the windings cause heat without producing real, useful power.
Before the breaker can be closed to parallel generators, four conditions need to be met between the oncoming generators and the bus to ensure smooth load transfer:
1) The oncoming generator should have the same or a slightly higher voltage than the bus.
2) The oncoming generator should have the same or a slightly higher frequency than the bus.
3) The phase angles need to match. For example, the oncoming generator “A” phase needs to be at 0 degrees when the bus “A” phase is at 0 degrees.
4) The phase sequences need to be the same. For example, A-B-C for the oncoming generator needs to match the A-B-C phase sequence of the bus.
Meeting these conditions can be visualized using Figure 1, which shows a time vs. voltage representation of an arbitrary, balanced three-phase signal. The bus and the generator each have their own corresponding plots resembling Figure 1, and the two should only be electrically connected if both plots line up and therefore satisfy the four conditions listed above.
If done properly, closing the breaker will be anticlimactic, and the gensets will happily find a new equilibrium. The gensets should be adjusted immediately to ensure the load is split evenly between gensets. If there is an electrical mismatch, the generator will instantly attempt to align its electrical phase with the bus, bringing the prime mover along for a wild ride and potentially causing physical damage—in addition to making a loud BANG! Idaho National Laboratories demonstrated the physical damage caused by electrical mismatch in its 2007 Aurora Generator Test.
Three primary setups for parallel genset operation are discussed here: droop-droop, isochronous-droop, and isochronous-isochronous.
The simplest mode of parallel operation between two or more gensets is a droop-droop mode, where both gensets are in droop mode and collectively find a new equilibrium frequency and voltage according to the real and reactive power demands of the load.
Isochronous-droop (iso-droop) mode is slightly more complex, where one genset is in droop mode and the other is in iso mode. The iso genset always provides the power required to maintain a specific voltage and frequency, and the droop genset produces a constant real power corresponding to that one point on its droop curve. Because the iso genset works more or less depending on the load, it is also termed the “swing” generator.
Finally, isochronous-isochronous (iso-iso) is the most complex. In iso-iso mode, both gensets attempt to maintain the specified output voltage and frequency. While this sounds ideal, this mode has the potential for instability during transient loading, because individual genset control systems may not be able to differentiate between a change in load and a change in the other genset’s power output. Iso-iso mode usually requires direct communication or a higher level controller to monitor both gensets, so they respond to load changes without fighting each other. With no external communication, one genset could end up supplying the majority of the power to the load while the second genset is idling, seeing no need to contribute because the bus voltage and frequency are spot on! At some point one genset could even resist the other genset, consuming real power and causing the generator to “motor” the prime mover. Unchecked, this condition will damage prime movers, so a reverse power relay is usually in place to trip the genset offline, leaving only one genset to supply the entire load.
Each genset simulated on the Hampden Training Bench had a custom sensor monitoring the generator voltage, current, and frequency output, a small computer running control calculations and a pair of DC-to-DC converters to close the control loop on the generator’s rotational velocity and field strength. The genset was simulated by coupling a 330 W brushed DC motor acting as the prime mover to a four-pole 330 W synchronous generator (Figure 2). Our power sensor was a custom-designed circuit board with an 8-bit microcontroller (MCU) employed to sample the genset output continuously and provide RMS voltage, RMS current, real power, reactive power, and frequency upon request. The control system ran on a Linux computer with custom software designed to poll the sensor for data, calculate the appropriate control response to return the system to the set point and generate corresponding pulse width modulated (PWM) outputs. Finally, the PWM outputs controlled the DC-to-DC converter to step down the DC supply voltage to drive the prime mover and energize the generator field coil. The component relationships are shown in Figure 3, where the diesel engine in a typical genset was replaced by our DC motor.
Since this project was a continuation of a previous year of work by Elise Sako and Jasper Campbell, several lessons were learned that required the system be redesigned from the ground up. One of the largest design constraint from the previous year was the decision to use a variable frequency drive (VFD) to drive an induction motor as the prime mover. While this solution is acceptable, it introduces inherent delay in the control loop, because the VFD is designed to execute commands as smoothly but not necessarily as quickly as possible.
Another design constraint was the decision to power the generator field coil using DC regulated by an off-the-shelf silicon controller rectifier (SCR) chopper. Again, while this is an acceptable solution, the system output suffered from the SCR’s slow response time (refresh rate is limited to the AC supply frequency), and voltage output regulation was non-ideal (capacitor voltage refresh again limited by the frequency of the AC supply).
To solve these performance constraints, we selected the responsive and easily controllable DC motor as the prime mover so the DC output from our Hampden Training Bench could be used as the power supply for both the DC motor and the generator field coil. By greatly simplifying the electrical control of the genset, we reduced implementation cost and improved control system response time.
The power sensor provided control feedback by taking continuous voltage and current measurements on a single phase of the generator output. Based on the design from the previous year, the sensor circuit was redesigned in Eagle CAD and printed. Ultimately the two-layer board routed just over 60 components on the top layer with an unbroken ground plane on the bottom, and measured 3″ by 3″. The circuit features a Microchip (formerly Atmel) ATmega328p MCU clocked at 20 MHz. Software libraries were developed for the Analog to Digital Converter (ADC), Inter-Integrated Circuit (I2C) and Universal Asynchronous Receive-Transmit (UART) MCU peripheral modules. A 2.5 kHz sampling rate was selected based on available clock prescalers for interrupt service routines, which gave a nearly perfect 42 samples per 60 Hz cycle. Voltage and current samples were managed with a circular buffer, and real-time RMS voltage, RMS current, real power, reactive power and frequency values for the connected phase were continuously calculated to be available immediately when data were requested.
Using the previous year’s design for reference, we refined the sensor schematic and hardware for sampling speed. To measure phase voltage, a single-phase and neutral (this configuration only works for a balanced “Y”-connected load) was fed into the primary coil of a small isolation transformer. The resulting voltage differential on the secondary of the transformer was treated as a complementary signal, and each path was attenuated independently using resistor dividers collectively biased around 2.5 V to allow the signal to occupy the full 5 V range of the ADC. ADC input filtering capacitors were selected according to the voltage divider impedance to low pass filter the signal at 1.7 kHz, rejecting noise at the ADC input pins. This corner frequency was selected because it avoided significant phase shift in the 60 Hz signal of interest, the phase shift of which is especially important for reactive power calculations.
To measure current, the generator output wire from the same phase was run through a current transformer loop. The secondary of the current transformer was loaded with a 100 Ω resistor, and the voltage difference, proportional to the current, was again treated as two complementary signals. After biasing around 2.5 V, the two signals were amplified using a rail-to-rail op amp in an instrumentation amplifier configuration to occupy the full 5 V range of the ADC. The final circuit board design shown in Figure 4 is rated to measure up to 340 VRMS and 12 ARMS on a single phase.
Several aspects of the C program were optimized to ensure faster sampling rates. Lookup tables were used for finding squares in real time, and resource-intensive square root and division calculations were offloaded to the Linux computer running our control system. RMS voltage was calculated using Equation 1, and RMS current was calculated using Equation 2 for our discrete series of voltage and current samples. Real and reactive power were calculated using Equations 3 and 4, respectively, which were adapted for discretely sampled signals.
Reactive power required a unique calculation, which delayed the voltage samples by 90 degrees from the current samples to isolate the purely reactive component. Forty-two samples per cycle is not perfectly divisible by 4, so a delay of 10 samples (instead of 10.5) was used for the reactive power calculation, and we incurred a continuous 4.3 degrees phase offset error.
To determine signal frequency, the two samples adjacent to the zero crossing were linearly interpolated to find the estimated zero crossing. The estimated time stamp of the previous zero crossing was subtracted from the present one, and the period was inverted to find frequency using the relationship shown in Equation 5. After the effective smoothing of the RMS calculations, the 42 samples per cycle produced excellent results, and sampling far above the Nyquist rate (of our primary 60 Hz signal) eliminated the need for signal reconstruction to determine the generator output frequency and voltage.
Because sampling loop speed was of primary importance, calculations were converted to a looping summation to ensure the values were always up-to-date when polled by the Linux computer. To accomplish this with minimal processing power, a moving window was used to identify the “active” samples in the circular buffer, so the program could simply add new samples to a running summation and remove the old samples.
Integer math was used for all summations to increase calculation speeds, which conveniently removed the possibility of error accumulation that might occur over time if using floating point math. When requested, the summation was sent to the Linux computer, which completed the divisions and/or square roots necessary, and scaled the ADC units to equivalent floating-point SI units. Scaling factors were initially determined mathematically, then adjusted empirically to ensure accuracy.
Curiously, our sensor sometimes produced RMS voltage readings greater than 1,000 V, or frequencies exactly half of the true system frequency. We had various hypotheses for these anomalies, and investigated edge cases in our integer summation windowing and zero-point crossing implementation, but were unable to resolve the spurious errors. The anomalies seemed to occur randomly, from multiple times per second to several seconds in between occurrences, so ultimately, we opted to remove the outliers with very specific filters. It was disappointing to not find the true solution, but we were forced to move forward due to time constraints.
The Linux computer running the control system had to bridge the gap between the low-level I2C data input, the PWM output and the higher-level network communication that would eventually be needed for multiple gensets to coordinate. Initially, we developed our control system to run on the low-cost and well documented Raspberry Pi. However, because the Raspberry Pi only generates a single PWM output, we ultimately switched to the BeagleBone development board, which had multiple onboard PWM signal generators. The BeagleBone is a similar Linux-based platform, so nearly all the code originally developed for the Raspberry Pi was simply compiled for the new architecture. The code developed for I2C communication with the sensor board remained the same, but a small library was written to control the new PWM hardware with a 10-bit duty cycle command.
After we nailed down our choice of silicon, we implemented the control system to regulate the system output at our desired 120 VRMS and 60 Hz. As the first step in any control system problem, we wanted to identify our system. Because of the perceived complication of implementing a true Multiple Input Multiple Output (MIMO) control system, however, we decided to make do with two Single Input Single Output (SISO) control systems running side by side. Ideally, frequency feedback would be used to control “throttle,” and the voltage feedback would be used to control the strength of the field coil electromagnet. The two parallel proportional and integral (PI) control systems were tuned separately to account for the slow mechanical time constant of the machine’s rotational speed and the faster electrical time constant of the rotor field strength.
Although it was easier for prototyping, the two SISO control systems were not truly independent from each; the generator output voltage and rotational speed affect one another. Another factor that was largely ignored during empirical tuning was the dynamic asymmetry in controlling rotational speed. It was easy to spin the machine up by increasing the throttle, but the machine could only slow down when coasting under load.
DC motors are known for their nearly linear relationship between terminal voltage and rotational speed (in steady state). As shown in Figure 5, an increase in applied DC motor terminal voltage “throttle” resulted in a linear increase in output frequency. Mathematically, we see the indicated proportional relationship between EMF and rotational speed (n) in Equation 6, because the motor’s speed constant (kn) and motor magnetic flux (Φ) are constants set during motor construction. Accounting for the voltage drop over the low-resistance motor windings, the relationship between terminal voltage and rotational speed are nearly linear. As the mechanically coupled generator is a synchronous machine, the electrical output frequency is approximately proportional to the terminal voltage on the DC motor.
Due to our haste to get the system up and running, we neglected to take measurements supporting the relationship between field coil voltage and generator output voltage, but we had experienced rough linearity when manually adjusting the machines. In lieu of data, we can analyze the governing equations to derive the expected response.
Inside a synchronous machine, the field coil is a spinning electromagnet, the current of which is proportional to the supplied voltage in accordance with Ohm’s Law. Equation 7 shows the Biot-Savart Law reduced to solve for the magnetic field (B) at the center of a single loop with a constant radius (r). Assuming that the permeability (μ) of the rotor, air gap, and stator are constant throughout the rotation, we see that the magnetic field of the rotor increases linearly with an increase in the current.
The derivative form of Faraday’s Law of induction in Equation 8 shows that EMF induced in a conductor is the change in magnetic flux, magnetic field (B) times area (A), over time. As differentiation is linear, any scaling of the magnetic field results in a linear scaling of the EMF experienced in the stator windings of a constant winding area. Therefore, as long as the generator’s core remains unsaturated, there should be a nearly linear relationship between the voltage applied to the field coil and the output voltage on the stator coils—again assuming the resistance of the stator windings are low, so voltage drop is minimal with increases in current.
After showing that the independent control systems of our genset should respond linearly, we ran our two PI control loops at 20 Hz to make constant corrections to the DC motor terminal voltage and the field settings. Armed with data from our power sensor, the controller calculated the error between the present state and the setpoint. For the proportional component, the controller multiplied the error by a fixed gain coefficient to determine the action needed to return the system to the set point, just like standard droop control. The integral component kept a running integral of the system’s error—the summation of error times loop period—and integrated error was multiplied by a gain coefficient and summed with the proportional component to determine the overall iso control response. Both the proportional and integral terms contributed to different aspects of the control system. Proportional response helped the genset react quickly, and integral response fine tuned the output to ensure steady state achieved the set point.
Because voltage was not produced on the genset output until the machine was spinning and no load could result in overspeed damage, we used a predetermined open loop start-up routine to ramp up speed and field voltage to 60% over the course of 5 seconds. Once the genset reached the end of its start-up routine, the control system took over. By delaying the integral term from accumulating until the generator control loop was closed, integral wind-up was avoided.
The final piece of equipment necessary to close the loop on our control system was the DC-to-DC converter to drive the DC motor and energize the generator field coil. For simplicity, we used a metal oxide semiconductor field effect transistor (MOSFET) as the power switch to “chop” the DC voltage from the Hampden Training Bench by rapidly turning the MOSFET on and off at the determined duty cycle. Several driver circuits and drive frequencies were tested before settling on a reliable design for the DC-to-DC converter. Ultimately, the open-collector of the opto-isolator receiving the PWM signal was used to drive the base of a PNP transistor in a common emitter configuration to supply the gate of our power MOSFET with the 12 V needed for saturation. Although this solution actively drove the MOSFET to conduct, the simplicity of the unipolar drive required a resistor to bleed the MOSFET gate current when switched off. All conduction waste heat was easily wicked away by the aluminum sheet to which the point-to-point circuits were mounted (Figure 6).
Working essentially as a buck converter using the inductive windings on the machine to smooth current from the DC power supply, several additions helped smooth the generator output and reduce switching harmonics. A snubber circuit was added across the MOSFET output to absorb switching energy from inductive spikes, and DC reactors were placed in line with the DC motor and generator field coil drivers to increase the effective inductance of our buck converter circuit. Body diodes of some spare insulated gate bipolar transistors (IGBTs) were used as freewheel diodes and prevented the MOSFETs from releasing their magic smoke. Despite these additions, a 1 kHz drive frequency (blue) still caused spikes in the generator output (green) shown in Figure 7. Non-ideal artifacts of the rapid charging and discharging of the DC reactor magnetic field were also evident in the distorted 60 Hz sinusoid, so the drive frequency was increased to 5 kHz.
Before using an opto-isolator as a buffer between the high-power driver circuitry and the Linux computer, high-voltage signals propagated back through the circuit and caused mysterious problems. Because we used the BeagleBone PWM hardware instead of a robust external PWM generator, energizing the genset caused the computer to go into a “kernel panic” state that halted all operations until it was reset. We spent several hours looking for errors in the I2C libraries on the MCU and the BeagleBone. Ultimately the use of opto-isolators electrically separated the controller from the power electronics, and snap-on ferrite chokes around the braided PWM signal lines prevented the electromagnetic interference (EMI) from disrupting the computer’s operation.
Because our objective was to maintain the system output at exactly 120 VRMS and 60 Hz, our original objective intentionally required iso-iso paralleling, which was a lofty goal. An iso-iso configuration posed instability challenges, with each individual genset control system fighting to manipulate the bus. Unfortunately, we ran out of time, and the software was never completed, but we spent our remaining time conceptualizing a master controller that would control an arbitrary number of gensets and balance a theoretically unlimited total power output.
The framework for the master controller was implemented in C on a laptop that communicated with the individual genset controllers via a TCP socket connection. After registering themselves with the master as online and available, the genset controllers entered paralleled mode, where they initiated connections with the master controller and sent periodic updates containing their RMS voltage, real power, reactive power and frequency.
Two methods were examined for master control: (1) cut out the genset controller and use the master controller to decide on genset-level control response; or (2) use the genset control systems for instant droop correction and use the master controller for long-term cooperative load balancing.
On the one hand, the first method would produce the best stability, because the master controller could respond to disturbances using the big picture to make decisions; however, pushing time-sensitive commands to a large number of gensets would be slow and would create a single point of failure. On the other hand, utilizing the genset controllers to respond instantly in droop mode would not fully return the bus frequency and voltage to the setpoint, but it would give the best overall stability even if the master controller failed. Long-term voltage, frequency and power adjustments could then be made by the master controller raising or lowering the droop curves for iso-iso control.
While fine tuning the long-term voltage and frequency, the master controller would simultaneously redistribute real and reactive power as necessary for maximum efficiency. Additionally, because the processing demand on the master controller would increase linearly with the control of more gensets, we considered this to be an optimal use of computing resources. Load distribution on the bus could be further tailored for maximum efficiency by identifying high efficiency “base load” gensets for normal conditions and lower efficiency “peaking” gensets when demand is high. Although it was never implemented, we were confident that our implementation would work well in island operation, especially with further control system tuning.
RESULTS AND DISCUSSION
Although we never finished the master controller, we successfully implemented a genset control system that exceeded our design requirements for steady-state operation and transient load recovery. Steady-state voltage and frequency output were maintained at 120 ±4 VRMS and 60 ±0.5 Hz. Two completed genset control systems were placed in an enclosure for operation and display (Figure 8).
Transient load response was examined by analyzing the recovery time for a stand-alone genset. Figure 9 shows the system step response to the switching-on of three “Y”-connected 60 W light bulbs—180 W is 55% of the genset power capacity. The system output dropped to roughly 75 VRMS at 49 Hz, but the control system was capable of returning the output to 120 VRMS at 60 Hz in about 2.5 seconds. Figure 10 shows the step response when the 180 W load was removed and the output jumped to 145 VRMS at 63 Hz. The output returned to steady state in about 3.5 seconds. The plots illustrate the control system’s less aggressive frequency correction to address the non-ideal dependence between frequency and voltage mentioned in the “Genset Controller” section.
Next, the control system was used to hold a genset in isochronous mode while it was paralleled against a second genset in droop mode. In paralleled island operation, a 180 W load had a much smaller effect. Although the tempting explanation is that because the overall system capacity is now 660 W, the droop genset only changes its output power when the bus frequency differs from 60 Hz. Therefore at 60 Hz the system capacity is limited to 330 W from the iso generator plus whatever power the droop generator is set to contribute.
The true benefit of this style of paralleling is that the droop genset improves the inertial response of the bus to transient loading, and also assists the iso genset by temporarily contributing more or less power to “bounce back” from large power fluctuations. Figure 11 and Figure 12 show that when the system was loaded and unloaded, the output only dropped to 103 VRMS at 54 Hz or rose to 150 VRMS at 62 Hz, respectively. Working as a swing generator, the controlled genset again responded immediately and met our objective by returning to steady state in about 3 seconds in both cases.
To demonstrate the significant increase in performance with our actively controlled genset, an 80 W induction motor (highly reactive load) was started and stopped using the same iso-droop configuration. During both tests, the 180 W resistive load remained continuously energized to provide a real component to the genset power output. As shown in Figure 13, with the control system, the iso-droop paralleled gensets temporary succumbed to the 90 V drop caused by the highly inductive load, but were able to spin up the motor’s inertia and fully recover in about 2.5 seconds (though a small voltage integral wind-up was visible in the aftermath). Figure 14 demonstrated the unloading of the control system, where the output rose by 30 V due to the removal of the inductive load; the system returned to steady-state in about 3.5 seconds.
While both induction motor tests show the performance of our voltage control system, they also highlight the dependence between the frequency and voltage control systems, since the majority of the disturbance was reactive. Ideally, the two exercises should have had a small impact on the frequency, aside from initially consuming real power to spin up the induction motor.
Successful hardware and software development of our genset control system demonstrated that a low-cost solution is attainable for managing gensets. Building and testing individual modules for any project involves component-level troubleshooting and further troubleshooting when integrating the various pieces to form the complete system. Despite the challenges and our imperfect PI controllers, the project demonstrated that an industrial control system can be designed and built completely from scratch and still have acceptable performance.
We surpassed our objective to keep steady-state output within ±VRMS and ±0.5 Hz, and nearly met our objective to respond to transient loads by returning to steady-state operation within 3 seconds. Even though the master controller was not completed, our discussions culminated in exciting realizations for stable iso-iso bus control schemes and provided a deeper appreciation for power systems in the industry and on board US Coast Guard cutters.
Properly identifying the mechanical and electrical characteristics of the genset would have greatly improved system response and provided stability during the testing of the master controller. For example, feed forward compensation might be able to decouple voltage and frequency response, or a tailored Multiple Input Multiple Output (MIMO) control system could potentially use this coupling to improve system response time and stability.
To completely automate the paralleling process, a “smart breaker” could also be developed to monitor the bus and oncoming genset, to determine when to make the connection. By measuring the differential voltage and beat frequency between both the “A” phase of the bus and oncoming genset, a device could detect compliance with the first three paralleling conditions. An additional device could be used on the “B” phase to validate the first three conditions and verify phase sequence—or this could be assumed for a properly installed circuit breaker.
Throughout the project and especially when developing the smart breaker, network vulnerability needs to be addressed to ensure machinery is protected from the physical damage that could result from malicious commands. Although the system would exist behind enterprise-level security appliances, in the event of network intrusion, machinery should be capable of recognizing legitimate authority or be smart enough not to cause damage directly.
For example, the gensets should know that the master would never command the system frequency be raised to 100 Hz, and the smart breaker should never allow itself to be overridden to connect the genset when it is 180 degrees out of phase with the bus. As a real example of the importance of security, the Ukraine power grid was disrupted in 2015 by phishing and malware attacks that caused extensive physical damage.
Regarding security in general, power distribution in the United States relies on synchrophasors, which use the low signal power global positioning system (GPS) constellation as a clock source to synchronize power grid waveforms across the country. As discussed in the Master Controller design section, gensets and the power grid can be vulnerable to instabilities, demonstrated by the cascading failures causing the 2003 Northeast Blackout, and systems need to be designed carefully to ensure robust operation.
For detailed article references and additional resources go to:
Microchip Technology | www.microchip.com
PUBLISHED IN CIRCUIT CELLAR MAGAZINE • FEBRUARY 2019 #343 – Get a PDF of the issueSponsor this Article
Kent Altobelli is from Los Angeles, California, and received his Bachelor of Science in Electrical Engineering from the U. S. Coast Guard Academy. After graduating as a Commissioned Officer in the U. S. Coast Guard, he served as a Deck Watch Officer on the Coast Guard Cutter ALDER out of Duluth, Minnesota. Kent is presently stationed in Miami, Florida supervising electronics support for Coast Guard units from South Carolina down to Puerto Rico.
Caleb Stewart is from Grandview, Texas, and received his Bachelor of Science in Electrical Engineering from the U.S. Coast Guard Academy. After graduation, he first served as a Deck Watch Officer onboard the Coast Guard Cutter BERTHOLF out of Alameda, California. Caleb is now the Infrastructure Support Branch Chief at the Coast Guard Cyber Security Operations Center. Caleb is an Associate Certified Information Systems Security Professional as well as an Offensive Security Certified Professional and is currently pursuing a Masters in Computer Science from the Georgia Institute of Technology.