High voltage microwave power filament voltage adjustment method

Abstract

The application discloses a high-voltage microwave power filament voltage adjustment method applied to a voltage adjustment circuit, and the method comprises the following steps: outputting a high-voltage side equivalent filament voltage according to a low-voltage side alternating current input voltage instantaneous amplitude, an alternating current output current amplitude and a transient adjustment signal; inputting the high-voltage side equivalent filament voltage and collected magnetron anode high-voltage direct current into a recursive time sequence model, so that the recursive time sequence model outputs a filament average reference temperature prediction value; determining a temperature minimum point of a filament axial direction according to the filament average reference temperature prediction value; determining whether to immediately release, delay execution or modify an inductance control instruction according to the temperature minimum point and a constraint judgment constant; and positioning a high-voltage side potential physical fault by using the alternating current output current amplitude, a duty cycle instruction of the transient adjustment signal, the high-voltage side equivalent filament voltage and a modulation amplitude index. Through the above method, the problem of transient stop vibration and high-voltage sparking caused by inductance switching delay and filament thermal inertia mismatch is solved.

Description

Technical Field

[0001] This application relates to the field of high-voltage microwave technology, and in particular to a method for adjusting the filament voltage of a high-voltage microwave power supply. Background Technology

[0002] In high-end microwave sources (such as industrial microwave power supplies driving thorium tungsten carbide cathode magnetrons), a "floating filament" control technique is typically used to extend the magnetron's lifespan. The basic principle is as follows: when the magnetron is first started, a high heating voltage is applied to the filament to excite thermionic electrons. Once the microwaves establish stable oscillations, the system relies on secondary electron emission generated by high-energy electrons bombarding the cathode to maintain the cathode temperature. At this point, a control circuit sends a command to activate an electronic switch or relay located on the high-voltage side (typically 3.0 to 5.5 kV), switching the power supply inductor on the high-voltage side from a single inductor to multiple inductors connected in series or parallel, thereby reducing the filament heating voltage. This method not only provides adequate energy to maintain the temperature but also prevents the filament from evaporating rapidly due to sustained high temperatures.

[0003] In the aforementioned existing technology scenario of "adjusting filament voltage by switching the high-voltage side inductor", there is a problem that seems simple in technology but is extremely difficult to solve completely in engineering: the problem of "transient oscillation stop" and "high-voltage arcing" caused by inductor switching delay and filament thermal inertia mismatch. Summary of the Invention

[0004] This application provides a method for adjusting the filament voltage of a high-voltage microwave power supply, which can solve the problems of transient oscillation stoppage and high-voltage arcing caused by inductor switching delay and filament thermal inertia mismatch.

[0005] In a first aspect, this application provides a method for adjusting the filament voltage of a high-voltage microwave power supply, applied to a voltage adjustment circuit. The voltage adjustment circuit includes: a hollow miniature transformer, employing a double-layer or multi-layer epoxy glass cloth copper-clad plate, with planar spiral coils drawn using copper foil traces on the top and bottom layers respectively, forming the primary and secondary windings of the hollow miniature transformer; a control circuit located on the low-voltage side of the hollow miniature transformer; a transmitting modulation circuit provided on the low-voltage side, connected to the control circuit; a receiving demodulation circuit provided on the high-voltage side of the hollow miniature transformer; a switching circuit, the control terminal of which is connected to the receiving demodulation circuit; and an inductive component, including at least one inductor, connected to the path terminal of the switching circuit, configured to, when the switching circuit is operating, use at least one inductor to connect the power supply inductor on the high-voltage side of the microwave power supply transformer in series or parallel to adjust the filament voltage; the switching circuit and the inductive component are located on the high-voltage side; the voltage adjustment method includes: the control circuit acquiring the instantaneous amplitude of the AC input voltage on the low-voltage side, and the primary circuit of the microwave power supply transformer... The control circuit takes the AC output current amplitude and the instantaneous adjustment signal output by the current proportional-integral regulator inside the control circuit as inputs. The control circuit infers the equivalent filament voltage on the high-voltage side based on the instantaneous amplitude of the low-voltage side AC input voltage, the AC output current amplitude, and the instantaneous adjustment signal. A recursive timing model is deployed within the control circuit. The control circuit inputs the equivalent filament voltage on the high-voltage side and the collected DC current of the magnetron anode high voltage to the recursive timing model, so that the recursive timing model outputs a predicted value for the average reference temperature of the filament. The control circuit determines the lowest temperature point along the filament axis based on the predicted value of the average reference temperature of the filament. The control circuit decides whether to immediately release, delay execution, or modify the inductor control command based on the lowest temperature point and the constraint judgment constant. The inductor control command is used to control the operation of the switching circuit. The constraint judgment constant is used to characterize the physical limit temperature required for the magnetron to maintain stable microwave output. When the switching circuit is operating, the control circuit uses the AC output current amplitude, the duty cycle command of the instantaneous adjustment signal, the equivalent filament voltage on the high-voltage side, and the modulation amplitude index to locate potential physical faults on the high-voltage side.

[0006] The control circuit includes a synchronous quadrature demodulation unit. The control circuit infers the equivalent filament voltage on the high-voltage side based on the instantaneous amplitude of the low-voltage side AC input voltage, the amplitude of the AC output current, and the output of an instantaneous adjustment signal. The process includes: the control circuit inputting the instantaneous adjustment signal to the synchronous quadrature demodulation unit, which multiplies the instantaneous adjustment signal with the grid synchronous quadrature signal generated by the phase-locked loop to obtain a first sequence and a second sequence; the control circuit inputting the first and second sequences into a first-order recursive digital low-pass filter to filter out the second harmonic component and extract the DC component, obtaining the first DC component corresponding to the first sequence and the second DC component corresponding to the second sequence; the control circuit obtaining the modulation amplitude index based on the first and second DC components; and the control circuit outputting the inferred equivalent filament voltage on the high-voltage side based on the modulation amplitude index, the instantaneous amplitude of the low-voltage side AC input voltage, the amplitude of the AC output current, the inherent turns ratio of the secondary and primary sides of the microwave power transformer, and the leakage inductance and equivalent line resistance of the primary side of the microwave power transformer.

[0007] The control circuit calculates the modulation amplitude index using the following formula: ; in, Indicates the first DC component. Indicates the second DC component. This indicates the modulation amplitude index.

[0008] When the recursive time series model is calibrated offline, the weight matrix corresponding to the recursive time series model satisfies the following linear matrix inequality: ;in, Denotes the first auxiliary matrix. Denotes the second auxiliary matrix. Describes a positive definite symmetric Lyapunov matrix. An augmented transition matrix that includes the system matrix and the boundaries of nonlinear terms.

[0009] The control circuit determines the lowest temperature point along the filament axis based on the predicted average reference temperature of the filament. This process includes: constructing a three-dimensional input tensor based on the predicted average reference temperature of the filament over the past K time steps, the spatial coordinates of the filament axis, and the inductor control command to be executed; performing frequency domain mapping and global integration on the three-dimensional input tensor to obtain the expected absolute temperature distribution curves at various points from the bottom to the top of the filament; and performing minimum value optimization on the expected absolute temperature distribution curves, using the minimum value as the lowest temperature point along the filament axis.

[0010] The control circuit performs frequency domain mapping and global integration on the three-dimensional input tensor to obtain the expected absolute temperature distribution curves of the filament from bottom to top. This includes: the control circuit projects the three-dimensional input tensor to a high-dimensional hidden feature space using a local linear transformation; the control circuit performs a discrete fast Fourier transform on the spatial domain features in the high-dimensional hidden feature space to obtain frequency domain features; the control circuit performs low-pass truncation on the frequency domain features and multiplies them with a pre-calibrated complex frequency domain weight matrix to obtain the inverse transform back to spatial domain features; and the control circuit performs dimensionality reduction on the inverse transform back to spatial domain features to obtain the expected absolute temperature distribution curves of the filament from bottom to top.

[0011] The control circuit determines whether to immediately release, delay execution, or modify the inductor control command based on the lowest temperature point and constraint judgment constant. This includes: the control circuit solving for the minimum intervention amount based on the lowest temperature point and constraint judgment constant; the control circuit issuing a new inductor control command in response to the existence of a minimum intervention amount and the minimum intervention amount being less than a first threshold; and the control circuit triggering a physical protection intervention strategy in response to the absence of a minimum intervention amount or the minimum intervention amount being greater than a second threshold, wherein the first threshold is less than the second threshold. The physical protection intervention strategy includes: the control circuit stopping the issuance of the new inductor control command; or, before issuing the new inductor control command, the control circuit controlling the anode high-voltage power supply loop to increase the anode set power before issuing the new inductor control command; or, the control circuit waiting for the minimum intervention amount to be solved in the next control cycle until the minimum intervention amount is less than the first threshold before issuing the new inductor control command.

[0012] The control circuit, when the switching circuit is operating, uses the AC output current amplitude, the duty cycle command of the instantaneous adjustment signal, the equivalent filament voltage on the high-voltage side, and the modulation amplitude index to locate potential physical faults on the high-voltage side. This includes: calculating a first residual based on the AC output current amplitude, the duty cycle command of the instantaneous adjustment signal, and the equivalent filament voltage on the high-voltage side; calculating a second residual based on the AC output current amplitude, the modulation amplitude index, and the equivalent filament voltage on the high-voltage side; determining that the switch contacts in the switching circuit are stuck due to sintering when the absolute value of the first residual is greater than a third threshold and the absolute value of the second residual is within a preset range of 0; and determining that the inductor in the inductive component has experienced an inter-turn insulation breakdown short circuit when the absolute value of the first residual is greater than the third threshold and the absolute value of the second residual is greater than the third threshold.

[0013] The method also includes: when the system is configured at the factory or dynamically adjusted, the control circuit uses the periodic disturbance time constant of the rated grid input of the microwave power supply, the upper limit amplitude of the DC bias voltage, the limit of the high-frequency AC disturbance amplitude, and the equivalent parasitic inductance and impedance parameters determined by the layout of the double-layer or multi-layer epoxy glass cloth copper-clad board to calculate the optimal capacitance value of the high-voltage side auxiliary power supply bypass absorption capacitor of the hollow micro transformer.

[0014] The method further includes: constructing a parameter proposal module and a physical compliance review module in the host computer; the host computer uses the parameter proposal module to generate a search subspace containing several candidate inductor combinations based on historical security configurations and boundary feedback summaries from the previous test; the host computer uses the physical compliance review module to trim the candidate inductor combinations in the search subspace to obtain a safe search subspace containing several target inductor combinations that meets the requirements; the host computer calls the underlying circuit to perform automated simulation of the target inductor combinations in the safe search subspace to obtain a simulation score for each target inductor combination; and selects a preset number of target inductor combinations from the simulation scores for power-on testing to obtain a real score; and determines the inductor combinations that need to be soldered and the optimal delay compensation time based on the simulation score and the real score.

[0015] The beneficial effects of this application are as follows: Unlike existing technologies, this application provides a high-voltage microwave power supply filament voltage adjustment method applied to a voltage adjustment circuit. The voltage adjustment circuit includes: a hollow miniature transformer, employing a double- or multi-layer epoxy glass cloth copper-clad plate, with planar spiral coils drawn using copper foil traces on the top and bottom layers to form the primary and secondary windings of the hollow miniature transformer; a control circuit located on the low-voltage side of the hollow miniature transformer; a transmitting modulation circuit is provided on the low-voltage side, connected to the control circuit; the hollow miniature transformer... The high-voltage side is equipped with a receiver demodulation circuit; a switching circuit, the control terminal of which is connected to the receiver demodulation circuit; and an inductive component, which includes at least one inductor, connected to the path terminal of the switching circuit. The inductive component is configured to, when the switching circuit is operating, connect the power supply inductor on the high-voltage side of the microwave power transformer in series or parallel using at least one inductor to adjust the power filament voltage. The switching circuit and the inductive component are located on the high-voltage side. The voltage adjustment method includes: the control circuit acquiring the instantaneous amplitude of the AC input voltage on the low-voltage side and the amplitude of the AC output current in the primary circuit of the microwave power transformer. The control circuit outputs the instantaneous adjustment signal from the proportional-integral current regulator inside the control circuit. Based on the instantaneous amplitude of the low-voltage side AC input voltage, the amplitude of the AC output current, and the instantaneous adjustment signal, the control circuit infers the equivalent filament voltage on the high-voltage side. A recursive timing model is deployed within the control circuit. The control circuit inputs the equivalent filament voltage on the high-voltage side and the acquired DC high-voltage current of the magnetron anode to the recursive timing model, so that the recursive timing model outputs a predicted average reference temperature value for the filament. The control circuit determines the lowest temperature point along the filament axis based on the predicted average reference temperature value. The control circuit determines whether to immediately release, delay execution, or modify the inductor control command based on the lowest temperature point and the constraint judgment constant. The inductor control command is used to control the operation of the switching circuit. The constraint judgment constant is used to characterize the physical limit temperature required for the magnetron to maintain stable microwave output. When the switching circuit is working, the control circuit uses the AC output current amplitude, the duty cycle command of the instantaneous adjustment signal, the equivalent filament voltage on the high-voltage side, and the modulation amplitude index to locate potential physical faults on the high-voltage side. This can solve the problems of transient oscillation stoppage and high-voltage arcing caused by inductor switching delay and filament thermal inertia mismatch. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of an embodiment of the voltage adjustment circuit provided in this application; Figure 2 yes Figure 1 A schematic diagram of the structure of one embodiment of the control circuit; Figure 3 This is a flowchart illustrating an embodiment of the high-voltage microwave power supply filament voltage adjustment method provided in this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It is understood that the specific embodiments described herein are only for explaining this application and not for limiting it. Furthermore, it should be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all structures. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0019] In high-end microwave sources (such as industrial microwave power supplies driving thorium tungsten carbide cathode magnetrons), a "floating filament" control technique is typically used to extend the magnetron's lifespan. The basic principle is as follows: when the magnetron is first started, a high heating voltage is applied to the filament to excite thermionic electrons. Once the microwaves establish stable oscillations, the system relies on secondary electron emission generated by high-energy electrons bombarding the cathode to maintain the cathode temperature. At this point, a control circuit sends a command to activate an electronic switch or relay located on the high-voltage side (typically 3.0 to 5.5 kV), switching the power supply inductor on the high-voltage side from a single inductor to multiple inductors connected in series or parallel, thereby reducing the filament heating voltage. This method not only provides adequate energy to maintain the temperature but also prevents the filament from evaporating rapidly due to sustained high temperatures.

[0020] In the aforementioned existing technology scenario of "adjusting filament voltage by switching the high-voltage side inductor", there is a problem that seems simple in technology but is extremely difficult to solve completely in engineering: the problem of "transient oscillation stop" and "high-voltage arcing" caused by inductor switching delay and filament thermal inertia mismatch.

[0021] Specifically, this manifests as follows: Because the control circuit is located on the low-voltage side, its control signal passes through a high-voltage isolation layer of several kilovolts (usually using a high-voltage optocoupler). Combined with the mechanical or physical action time of the high-voltage electronic switch itself, there is an inherent action delay (microseconds to milliseconds). Within this delay window after the system issues the "increase inductance, decrease voltage" command, if the microwave load fluctuates slightly, the heat dissipation rate of the cathode filament will momentarily exceed the heat generation rate. Since this cannot be predicted in advance, by the time the inductor actually completes the switching, the filament temperature has often already fallen below the minimum physical threshold for maintaining secondary electron emission, directly causing the microwave source to experience a "transient oscillation stop." Simultaneously, abruptly switching the inductor under high voltage conditions can easily trigger strong high-frequency oscillations of parasitic capacitances in the circuit, generating unpredictable voltage spikes that can break down the auxiliary power supply providing energy to the switch, leading to high-voltage arcing.

[0022] This application focuses on solving the following problem: how to accurately predict and compensate for the filament temperature drop during the delay period in a high-voltage environment with inherent signal transmission delay and high-frequency transient oscillation, so as to achieve lossless and smooth switching of the inductance on the high-voltage side and ensure that the cathode filament temperature is always maintained above the emission threshold.

[0023] In some embodiments, such as Figure 1 As shown, the same transformer (half-bridge or full-bridge circuit structure driving the transformer) generates the anode high voltage and the cathode filament voltage. PWM or PFM technology is used to regulate the power supply's output voltage. Because the same transformer structure is used, only different secondary windings are used to generate the corresponding anode high voltage and cathode filament voltage. Both secondary windings are wound on the same secondary magnetic core, so the voltages have a strong correlation. That is, the anode high voltage and the cathode filament voltage are corresponding, and since the anode high voltage is fixed, the cathode filament voltage is also fixed. Because the cathode filament (thorium tungsten carbide cathode) is heated to approximately 1500℃-600℃ when the microwave magnetron is working (the addition of thorium and carbide treatment significantly reduces the operating temperature, improves electron emission efficiency and filament life, and is currently the most widely used type).

[0024] The following is a brief description of the magnetron startup and operation process: A magnetron is a cleverly designed device that can efficiently convert electrical energy into high-power microwaves without requiring complex semiconductor amplifier circuits. A magnetron mainly consists of three core components: a cathode, an anode, and a magnet.

[0025] Cathode: Located in the center, it is heated when energized and is responsible for emitting electrons.

[0026] Anode: Surrounding the cathode, it is usually a copper block with multiple resonant cavities. The resonant cavity is the "container" for microwave oscillation.

[0027] Magnet: It generates a strong magnetic field parallel to the axis, penetrating the space between the cathode and anode.

[0028] The startup process is like turning a group of disordered electrons into an organized "spoke" array.

[0029] Electrons start: First, the cathode is heated by the filament, and a large number of electrons "boil" due to the heat and escape from the cathode surface, forming an electron cloud surrounding the cathode.

[0030] Electric field acceleration: When a high DC voltage is applied between the cathode and anode (anode positive, cathode negative), the strong electric field pulls electrons radially from the cathode toward the anode.

[0031] Magnetic field deflection: Without a magnetic field, electrons would fly in a straight line to the anode, forming a simple current. However, because of the presence of a strong axial magnetic field, everything changes. According to the Lorentz force law in physics, a moving electron in a magnetic field experiences a lateral force. This force causes the electron, which was originally flying in a straight line, to no longer rush towards the anode, but instead rotate around the cathode along a curve. Operating mechanism: Energy conversion from "rotation" to "oscillation". Once electrons begin to rotate around the cathode, a more crucial stage begins.

[0032] The formation of "electron spokes": The rotating electron cloud is not uniform. When electrons pass through the gaps in the anode resonant cavity, they generate a weak high-frequency electromagnetic field within the cavity. This high-frequency field, in turn, "combs" the electron cloud, pushing, pulling, and gathering the electrons, ultimately forming a rotating group of electrons like spokes. This is the famous "electron spokes".

[0033] Maintaining oscillation (energy exchange): This is the most crucial step in the magnetron process. When the "electron spokes" sweep across the anode surface at a specific speed, electrons that happen to be in the decelerating electric field region transfer some of their kinetic energy to the high-frequency field, thus maintaining and enhancing the microwave oscillations in the resonant cavity. This process can be imagined as pushing a swing at the right time, making it swing higher and higher.

[0034] Energy output: Once the microwave oscillations in the resonant cavity accumulate to a certain level, the microwave energy can be guided out through structures such as antennas or coupling loops for external use.

[0035] In summary, the magnetron cleverly utilizes orthogonal electric and magnetic fields to continuously and efficiently pump the kinetic energy of electrons into a microwave electromagnetic field, thereby generating high-power oscillations. Once activated, the entire system achieves a self-sufficient dynamic equilibrium. Based on the above, a phenomenon was discovered: after the magnetron is started, the cathode filament does not need continuous high-power heating. Sometimes, even when the heating is disconnected, microwaves can still be output, but after a long time, they will stop emitting waves. This revealed a problem: if the heating voltage of the filament can be reduced to continue replenishing energy, but only the temperature of the filament needs to be maintained, it is not possible to heat it too high, because the filament aging will be accelerated, which will seriously affect its lifespan.

[0036] The following explanation will be based on both physical principles and engineering practice: Even after the heating is disconnected after startup, the microwaves can still output microwaves for a period of time. This is due to the switching between two electron emission mechanisms: Start-up phase: Relies on thermionic emission. When the magnetron first starts working, the cathode is cold and contains no electrons. A large current must be supplied to the filament to heat the cathode to 1000-1600℃ (depending on the material), allowing the electrons to gain enough energy to "boil" out. This heating power is necessary in this stage to enable the cathode to emit electrons.

[0037] Operational Phase: Relies on secondary electron emission. Once the oscillation is established, high-speed electrons begin bombarding the cathode surface. This bombardment produces a physical phenomenon: secondary electron emission. A single high-speed electron colliding with the cathode may eject several electrons. When the magnetron enters a stable oscillation state, the number of secondary electrons generated by the electron bombardment is sufficient to maintain the electron cloud density within the tube.

[0038] Based on this, the cathode is "heated" by electrons themselves and "collides" to produce enough electrons, enabling it to work even when the heating is disconnected.

[0039] Although electron bombardment generates secondary electrons and heat, if the heat generated is insufficient to compensate for the radiative and conductive heat loss from the cathode surface to the environment, the cathode temperature will slowly decrease. When the cathode temperature drops below a certain threshold, the thermionic emission capability weakens significantly. Once thermionic emission is insufficient, the number of electrons bombarding the cathode decreases, leading to a decline in secondary electron yield and creating a vicious cycle. When the electron cloud density falls below the minimum density required to maintain oscillation, microwave output will stop. Simply put: the cathode is slowly cooling down until it can no longer emit electrons. This application proposes reducing the heating voltage of the filament while continuing to provide energy, only maintaining the temperature. The following points need to be considered from an engineering perspective for this technology: The optimal operating temperature setting should not be set too high, as it would be at startup (that would severely accelerate filament evaporation and shorten lifespan). Nor should it be set too low, otherwise it will reach the aforementioned "collapse" critical point.

[0040] Ideally, a balance point should be found where the filament temperature just compensates for heat loss, ensuring that the cathode still has sufficient electron emission capability even when electron bombardment is insufficient (such as when microwave power is reduced).

[0041] Power and heating linkage control: If the magnetron output power is very high (meaning intense electron bombardment), the filament heating can be significantly reduced or even completely cut off (by self-sustaining).

[0042] If the output power is very low (light load) and the electron bombardment is weak, it is necessary to increase the filament's sustaining heating power to prevent the cathode from cooling down.

[0043] High-end switching power supplies monitor output power or current in real time and dynamically adjust filament voltage.

[0044] The essence of filament aging: The main causes of filament aging are evaporation at high temperatures (thinning of the tungsten filament) and the consumption of the cathode coating. If the temperature can be precisely controlled during continuous wave operation, keeping the cathode just above the "emission threshold," the lifespan of the magnetron can indeed be significantly extended. This is a crucial technique in industrial microwave sources that require continuous operation for tens of thousands of hours (such as plasma excitation and microwave drying).

[0045] In engineering, this is usually solved in the following ways: Traditional transformer power supplies typically employ a simple and crude method of "high heat first, then low heat" or simply disconnecting the power supply (because the transformer's turns ratio is fixed, making precise control difficult). Many household microwave ovens use a simplified logic of cutting off the filament after startup, hoping that the cathode temperature won't drop significantly during a single operating time (a few minutes).

[0046] Switching inverter power supplies can precisely control the filament voltage. They provide full power during startup (e.g., 3.15V) and then reduce the voltage to a "holding voltage" (e.g., 1.5V ~ 2.0V) after startup. This maintains the cathode temperature and prevents high-temperature evaporation, thus extending the lifespan of the lamps.

[0047] Based on this, this application adjusts the voltage of the cathode filament by adjusting the inductance value of L102 (L102 is replaced by multiple inductors connected in series and parallel). The application also adds switching switches RL1 and RL2 (the types of switches include electronic switches such as relays, solid-state relays, MOSFETs, IGBTs, thyristors, and diodes) and the drive circuits for the switches.

[0048] When the magnetron first starts up, L102 is an inductor, and the output cathode filament voltage is relatively high, which rapidly heats the cathode filament. When the microwave reaches a stable state (about 1 minute and 30 seconds), the MCU (control circuit) monitors the anode current, anode high voltage, transformer power, etc. When it is necessary to reduce the filament voltage, the MCU sends a command to the switch drive circuit to change the inductance value of L102 to adjust the filament voltage.

[0049] For example, replacing the L102 inductor with L103 and L104 connected in series makes it easier to change the inductance of L102. Switching will cause a change in the inductance when the switch is turned on (short circuit or open circuit).

[0050] Adjusting the inductance of L102 is equivalent to adjusting the energy stored in the inductor, which in turn changes the voltage of the cathode filament.

[0051] Furthermore, since the cathode filament voltage is superimposed on the anode high voltage, which reaches 3.0-5.5KV, a switching switch and corresponding drive circuit need to be designed for this high voltage. Insulation and auxiliary power supply present significant engineering challenges. Therefore, this application uses a switching switch (the switch type includes relays, solid-state relays, MOSFETs, IGBTs, thyristors, diodes, and other electronic switches) to adjust the energy storage inductance at the cathode filament, thereby adjusting the cathode filament voltage. The energy storage inductance at the cathode filament is adjusted using one or more inductors connected in series and parallel (one inductor is tapped). It also includes a circuit that directly uses the cathode filament winding for rectification and filtering to provide auxiliary power to the switching switch. Figure 1 D18, D19, and EC1 are for rectification and filtering, and the auxiliary power supply is superimposed on the anode high voltage.

[0052] The voltage regulation circuit can include time control, microwave stability monitoring, control of the cathode filament voltage versus anode current curve, anode current monitoring, etc., or it can be directly controlled by sending commands from the MCU.

[0053] Figure 2 U2 is the driver circuit control IC (control circuit), and Q8 is the driver enhancement transistor.

[0054] This part of the switching, inductor, and drive circuit is housed in a highly insulating plastic casing and treated with insulating glue to solve the insulation problem.

[0055] In some embodiments, a high-insulation relay (such as an insulation relay of 5KV or above) can be used as a switching switch, thus eliminating the need to draw power directly from the cathode filament winding for auxiliary power supply. However, inductance switching, switching, inter-station control, microwave stability monitoring, cathode filament voltage versus anode current curve control, and anode current monitoring are all still necessary.

[0056] Combination Figure 1 and Figure 2 The basic circuit structure of the voltage regulation circuit is explained below: The voltage regulation circuit includes: a hollow miniature transformer, a control circuit, a switching circuit, and an inductive component.

[0057] The hollow micro transformer uses double or multi-layer epoxy glass cloth copper-clad plates, and uses copper foil traces to draw planar spiral coils on the top and bottom layers to form the primary winding and secondary winding of the hollow micro transformer.

[0058] The control circuit is located on the low-voltage side of the hollow miniature transformer; a transmitting modulation circuit is installed on the low-voltage side, and this transmitting modulation circuit is connected to the control circuit. A receiving demodulation circuit is installed on the high-voltage side of the hollow miniature transformer. The control circuit can be composed of a microcontroller and peripheral circuits. The microcontroller can be, for example... Figure 2 U2 in the middle.

[0059] The control terminal of the switching circuit is connected to the demodulation circuit of the receiving terminal. The switching circuit is as follows: Figure 1 RL1 and RL2 in the example.

[0060] The inductive component includes at least one inductor connected to the path of the switching circuit. It is configured to, when the switching circuit is operating, connect the power supply inductor on the high-voltage side of the microwave power transformer in series or parallel using at least one inductor to adjust the power filament voltage. The switching circuit and the inductive component are located on the high-voltage side. The inductor is as follows: Figure 1 L103 and L104 in the example.

[0061] See Figure 3 , Figure 3 This is a schematic flowchart of an embodiment of the high-voltage microwave power supply filament voltage adjustment method provided in this application. It is applied to the voltage adjustment circuit mentioned in this application. The voltage adjustment method includes: Step 31: The control circuit acquires the instantaneous amplitude of the AC input voltage on the low-voltage side, the amplitude of the AC output current of the primary circuit of the microwave power transformer, and the instantaneous adjustment signal output by the current proportional-integral regulator inside the control circuit.

[0062] In some embodiments, in existing high-voltage microwave power supplies (operating voltage typically 3.0 to 5.5 kV), high-voltage optocouplers are usually used for signal isolation between the low-voltage control side and the high-voltage execution side, and a high-voltage divider resistor network is used for voltage status feedback. However, in actual production scenarios, optocouplers suffer from severe microsecond-level signal transmission delays, causing inductor switching to lag behind filament temperature changes; simultaneously, high-voltage divider resistors are prone to temperature drift or even local breakdown under long-term high-frequency microwave electromagnetic fields, becoming a weak link that can cause high-voltage arcing.

[0063] This solution abandons the use of dedicated high-voltage integrated isolation chips and instead uses fully discrete general-purpose semiconductor devices and printed circuit board (PCB) traces to construct the isolation driver hardware.

[0064] Physical isolation structure: Double or multi-layer epoxy fiberglass cloth copper-clad laminate (e.g., conventional 1.6 mm thick FR-4 board) is used. Planar helical coils are drawn using copper foil traces on the top and bottom layers, forming the primary and secondary windings of the hollow miniature transformer. Utilizing the inherent dielectric strength of FR-4 material (greater than 20 kV / mm), electrical isolation of greater than 3 kV or even 5 kV can be achieved with a physical spacing of only 1.6 mm without any additional potting compound or insulating components.

[0065] Transmitter modulation circuit (low-voltage control side): On the primary winding side of the hollow miniature transformer, two cross-coupled N-channel insulated-gate field-effect transistors and the PCB primary winding together form a self-excited inductor-capacitor (LC) high-frequency oscillator. When the control circuit needs to issue an "inductor switching" command, it outputs a standard 5-volt logic high level to power the oscillator, and the oscillator instantly starts oscillating (operating frequency set between 15 MHz and 30 MHz). The low-frequency logic control signal is modulated into high-frequency electromagnetic energy through on / off keying (OOK), and coupled to the secondary winding through the PCB substrate.

[0066] Demodulation circuit at the receiver (high-voltage execution side): On the secondary winding side of the hollow miniature transformer, two general-purpose NPN bipolar transistors and their bias resistors constitute a full-wave rectification and amplification circuit. When the secondary winding is induced with a high-frequency AC voltage, the bipolar transistor turns on and enters the saturation region, outputting a low level (logic 0). Its ingenious core lies in utilizing the inherent recovery time of the bipolar transistor exiting the saturation region (typically around 100 nanoseconds), allowing it to naturally act as a hardware low-pass filter, filtering out the 15 MHz high-frequency carrier and directly restoring a stable and smooth control level to directly drive the high-voltage side switching circuit (such as a silicon carbide MOSFET or a high-voltage solid-state relay).

[0067] Physical effect: The signal propagation delay of this pure hardware structure is strictly controlled within 200 nanoseconds, which is more than an order of magnitude lower than that of traditional optocouplers, providing an extremely accurate action time reference for subsequent delay-free temperature field prediction and control.

[0068] In some embodiments, the control circuit acquires the instantaneous amplitude of the low-voltage side AC input voltage in real time. The amplitude of the AC output current in the primary circuit of the microwave power transformer And the instantaneous adjustment signal output by the proportional-integral (PI) regulator inside the control circuit. .

[0069] Step 32: The control circuit infers the equivalent filament voltage on the high-voltage side based on the instantaneous amplitude of the low-voltage side AC input voltage, the amplitude of the AC output current, and the instantaneous adjustment signal output.

[0070] In some embodiments, the control circuit includes a synchronous quadrature demodulation unit, and step 32 may be the following process: Step 321: The control circuit inputs the instantaneous adjustment signal to the synchronous quadrature demodulation unit, so that the synchronous quadrature demodulation unit multiplies the instantaneous adjustment signal with the grid synchronous quadrature signal generated by the phase-locked loop to obtain the first sequence and the second sequence.

[0071] To infer the high-voltage side state, the modulation depth (modulation amplitude index) of the current power supply needs to be obtained first. Since the AC signal changes rapidly over time, this algorithm constructs a synchronous quadrature demodulation (I / Q demodulation) unit within the control circuit: The instantaneous adjustment signal of the PI controller output. Multiply by the power grid synchronization orthogonal signal generated by the phase-locked loop respectively: ; ; in, For discrete sampling periods, Angular frequency, This represents the phase angle of the power grid.

[0072] in, Indicates the first sequence, This indicates the second sequence.

[0073] Step 322: The control circuit inputs the first sequence and the second sequence into a first-order recursive digital low-pass filter to filter out the second harmonic component and extract the DC component, thereby obtaining the first DC component corresponding to the first sequence and the second DC component corresponding to the second sequence.

[0074] The multiplied sequence is then input into a first-order recursive digital low-pass filter to filter out the second harmonic component and extract the DC component: ; ; in , This is the filter cutoff frequency, set according to the power frequency, used to balance noise suppression and dynamic response.

[0075] in, Indicates the first DC component. This represents the second DC component.

[0076] Step 323: The control circuit obtains the modulation amplitude index based on the first DC component and the second DC component.

[0077] The control circuit calculates the modulation amplitude index using the following formula: ; in, Indicates the first DC component. Indicates the second DC component. This indicates the modulation amplitude index.

[0078] Step 324: The control circuit outputs the inferred equivalent filament voltage on the high-voltage side based on the modulation amplitude index, the instantaneous amplitude of the low-voltage side AC input voltage, the amplitude of the AC output current, the inherent turns ratio of the secondary and primary sides of the microwave power transformer, and the leakage inductance and equivalent line resistance of the corresponding primary side of the microwave power transformer.

[0079] During steady-state and quasi-steady-state operation of the microwave power supply, the primary low-voltage side and the secondary high-voltage side (filament power supply side) strictly follow Kirchhoff's voltage law and transformer coupling equations. This step will extract the modulation amplitude index. Treating it as a bridge connecting the primary and secondary sides, the following physical equations are constructed: .

[0080] in, The inferred amplitude of the equivalent fundamental voltage at the high-voltage side filament port (equivalent filament voltage on the high-voltage side); and These are the leakage inductance and equivalent line resistance of the transformer referred to the primary side (the leakage inductance and equivalent line resistance of the primary side of the microwave power supply transformer) (which are known constants). This represents the inherent turns ratio between the secondary and primary sides of a high-voltage transformer.

[0081] The final output is the real-time inferred equivalent filament voltage on the high-voltage side. This output data directly reflects the actual voltage division effect of the high-voltage side inductor network (whether it is a single inductor or multiple inductors connected in series). The control circuit compares this inferred value (equivalent filament voltage on the high-voltage side) with the preset theoretical voltage range for each inductor level. This allows for precise confirmation, without any physical sensors connected to the high-voltage side, of whether the electronic switch (switching circuit) on the high-voltage side has successfully performed the switching action, and the actual voltage level currently applied to the filament.

[0082] That is, the control circuit performs real-time extraction of the modulation amplitude index and high-voltage state mapping based on the electromagnetic coupling physical equation in the above process, and obtains the equivalent filament voltage on the high-voltage side.

[0083] By streamlining the process as described above, this step achieves accurate sensing of the dynamic state of the 3.0-5.5 kV high-voltage system without adding any complex high-voltage side measurement components. It has the following effects: To effectively avoid the defect of "algorithms having no actual physical meaning", the orthogonal demodulation and mapping algorithm in this step has all its inputs and outputs closely bound to the real physical quantities (voltage, current, leakage inductance) of the microwave power supply. The mathematical derivation is entirely based on the objective law of electromagnetic coupling of transformers, and has extremely strong industrial feasibility.

[0084] The design is remarkably unconventional and innovative: the conventional design approach in this field is to "add sensors wherever data is needed," but this solution takes the opposite approach, utilizing the internal parameters (modulation amplitude index) of the digital regulator on the low-voltage side to perform implicit physical back-calculation across the isolation layer. This not only enables absolute insulation safety on the high-voltage side with "zero sensors," but also makes it possible to use a feedback-free, extremely low-latency PCB micro-transformer as a pure downlink isolation driver, fundamentally solving the problem of the difficulty in balancing high-voltage arcing and signal delay.

[0085] Step 33: The control circuit inputs the equivalent filament voltage on the high-voltage side and the collected DC current of the magnetron anode to the recursive timing model so that the recursive timing model outputs the predicted value of the average reference temperature of the filament.

[0086] The control circuit incorporates a recursive timing model.

[0087] When the recursive time series model is calibrated offline, the weight matrix corresponding to the recursive time series model satisfies the following linear matrix inequality: ;in, Denotes the first auxiliary matrix. Denotes the second auxiliary matrix. Describes a positive definite symmetric Lyapunov matrix. An augmented transition matrix that includes the system matrix and the boundaries of nonlinear terms.

[0088] In continuous wave operation of industrial magnetrons, the temperature distribution of the cathode filament is not a uniform single point, but rather exhibits a temperature gradient along the axial direction (the ends are colder due to heat conduction, while the middle section is hotter due to electron backfire). Within the millisecond-level delay window of high-voltage inductor switching and a sudden drop in filament heating voltage, the "coldest points" at both ends of the filament are very likely to be the first to fall below the physical threshold for secondary electron emission (e.g., 1500 degrees Celsius), thus triggering a "transient oscillation stop" of the microwave source.

[0089] Traditional prediction methods based on a single temperature measurement point or a coarse thermal resistance-capacitance model cannot solve for the evolution of the entire spatial temperature field within a microsecond-level computation cycle, nor can they guarantee the convergence of mathematical iterations under strong electromagnetic disturbances. This step provides a method that can deduce the one-dimensional spatial temperature distribution field on the surface of a cathode filament in real time within a future time window without requiring massive computing power, accurately pinpointing the "coldest point" for control decisions.

[0090] Phase 1: Constructing a recursive time series model constrained by the physical energy conservation principle (deriving the average reference temperature of the filament).

[0091] This phase aims to address the problem that traditional data-driven models tend to output divergent data that "violates thermodynamic principles" under strongly nonlinear conditions.

[0092] Input data settings (input to the recursive time series model): The model at the current sampling time input vector It includes two physical quantities: one is the equivalent filament voltage on the high-voltage side obtained by the sensorless inference mentioned above. (Characterizing the Joule heating power input); secondly, the high-voltage DC current of the magnetron anode synchronously sampled by the control circuit. Characterizing the secondary heating effect generated by high-speed electron backflushing of the cathode.

[0093] Physical intrinsic connections and equations of state: Define the system state vector This represents the baseline internal thermal state of the cathode. Based on the first law of thermodynamics (Joule heating + electron bombardment heat = heat conduction loss + heat radiation loss), a discrete-time recursive state equation with a hidden layer is constructed: ; ; in, The weight matrix to be calibrated For non-linear activation functions (such as the hyperbolic tangent function), the output is... For a moment The predicted average reference temperature of the filament.

[0094] Stability hard constraints based on linear matrix inequalities (key to overcoming algorithm stacking): To ensure that the equation does not output a divergent temperature that violates physical laws under any extreme voltage fluctuations, the conventional unconstrained gradient descent is not used during the offline calibration of the model (solving the weight matrix). Instead, input-to-state stability constraints are forcibly injected.

[0095] The specific operation is as follows: Solve for a positive definite symmetric Lyapunov matrix. and related auxiliary matrices , It is mandatory that all weight matrices must strictly satisfy the following linear matrix inequality (LMI) during the optimization process: ; (in For including system matrix And the augmented transition matrix of the nonlinear term boundary).

[0096] The physical significance of the aforementioned matrix inequality lies in the fact that it mathematically establishes that the system's energy dissipation rate is always greater than its energy accumulation rate, thus guaranteeing the derived predicted average reference temperature of the filament. When faced with a step-like inductor switching disturbance, it will inevitably converge to a new physical steady state, completely eliminating the risk of "numerical explosion" in pure algorithm models.

[0097] Step 34: The control circuit determines the lowest temperature point along the filament axis based on the predicted average reference temperature of the filament.

[0098] In some embodiments, step 34 may be the following process: Step 341: The control circuit constructs a three-dimensional input tensor based on the predicted average reference temperature of the filament over the past K time steps, the spatial coordinates of the filament axis, and the inductor control command to be executed.

[0099] Step 342: The control circuit performs frequency domain mapping and global integration on the three-dimensional input tensor to obtain the expected absolute temperature distribution curves of the filament from the bottom to the top.

[0100] In some embodiments, the control circuit projects the three-dimensional input tensor to a high-dimensional hidden feature space using a local linear transformation; the control circuit performs a discrete fast Fourier transform on the spatial domain features in the high-dimensional hidden feature space to obtain frequency domain features; the control circuit performs low-pass truncation on the frequency domain features and multiplies them with a pre-calibrated complex frequency domain weight matrix to obtain inverse transform back spatial domain features; the control circuit performs dimensionality reduction on the inverse transform back spatial domain features to obtain the expected absolute temperature distribution curves of the filament from bottom to top.

[0101] Step 343: The control circuit performs minimum value optimization on the expected absolute temperature distribution curve and takes the minimum value as the lowest temperature point along the filament axis.

[0102] In some embodiments, stage two: spatial temperature field analysis based on frequency domain global mapping (deriving one-dimensional spatial temperature gradient) After obtaining the constrained reference temperature time series (predicted average reference temperature of the filament), it must be extended to the spatial dimension to find the lowest temperature point along the filament axis. Limited by the computing power of the control circuit, it is impossible to solve the complex partial differential heat conduction equations (partial differential equations) online. This stage employs a method of mapping spatial variables to the frequency domain for global solution.

[0103] The input data tensor quantization settings are as follows: Construct a multi-channel 3D input tensor The channels of this tensor include: past... Reference temperature sequence at each time step (Predicted average reference temperature of the filament), spatial coordinate channel (representing the axial discrete physical coordinate sequence of the filament from bottom to top). , filament axial spatial coordinates), and the sequence of high-voltage inductor switching action instructions to be executed (inductor control instructions, i.e., control action constant channels).

[0104] Frequency domain mapping and global integral operators (substantial improvements to replace complex calculations): High-dimensional feature projection: First, the input tensor is transformed using local linear transformation. Projected to a higher-dimensional hidden feature space .

[0105] Fast frequency domain convolution solution: Integral kernel mapping is performed in the hidden layer. Unlike traditional point-by-point difference calculation of heat conduction between adjacent nodes, this step applies Discrete Fast Fourier Transform (FFT) to the spatial coordinates to transform the spatial domain features. Convert to frequency domain .

[0106] Low-pass cutoff multiplication with the frequency domain: In the frequency domain, the smooth global characteristics of heat transfer are characterized by a concentration of low-frequency components. Therefore, a cutoff wavenumber is set. Only retain frequencies with absolute values ​​less than 100%. The low-frequency Fourier mode, and compared with a pre-calibrated complex frequency domain weight matrix. Multiplication: ; This step is mathematically equivalent to instantly performing a Green's function integral of the global heat conduction equation over the entire filament length. Its computational complexity is only one-thousandth of that of the traditional finite element differential algorithm, perfectly suited to the scenario where the computing power of the control circuit is limited in actual production.

[0107] Spatial field output reconstruction: The features transformed back to the spatial domain are subjected to dimensionality reduction through a linear layer, and the final output objective function is obtained. .

[0108] Output data settings and their physical meaning: Output data It is a spatial discrete vector that precisely represents the system's inherent delay time from the issuance of the inductor switching command. Then, the expected absolute temperature distribution curves of the filament from bottom to top. .

[0109] Phase 3: Extreme value extraction and control variable generation (interface association).

[0110] After completing the frequency domain spatial mapping, the microcontroller outputs a one-dimensional temperature distribution curve within the current prediction period. Perform a minimum value search operation: .

[0111] This minimum value This represents the expected "coldest point" temperature at which the cathode filament surface is most likely to lose its secondary electron emission capability due to cooling at the end of the upcoming delay window.

[0112] This output data It will be directly used as the sole feedforward decision benchmark input in step 35.

[0113] This step is an organic combination of strict matrix inequality physical constraints (to solve the time-series convergence problem) and frequency domain integral operator mapping (to solve the problem of dimensionality reduction of spatial partial differential equations), rather than a simple patchwork of conventional algorithms.

[0114] Without adding any physical temperature sensors or relying on high-performance graphics processing units (GPUs), it implements a complete "thermal-electrical coupled spatiotemporal digital twin" within conventional industrial control circuits. This implementation method establishes a physical data flow from "low-level electrical parameter inference" to "future spatial temperature field distribution," providing a highly physically interpretable and fully implementable core sensing foundation for solving the "transient vibration cessation" during inductor switching.

[0115] Step 35: The control circuit determines whether to immediately release, delay execution, or modify the inductor control command based on the lowest temperature point and the constraint judgment constant.

[0116] The inductor control command is used to control the operation of the switching circuit; the constraint judgment constant is used to characterize the physical limit temperature required for the magnetron to maintain stable microwave output.

[0117] In some embodiments, the control circuit solves for the minimum intervention amount based on the lowest temperature point and constraint determination constants.

[0118] The control circuit responds to the presence of a minimum intervention amount, which is less than a first threshold, by using a downward induction control command; the control circuit responds to the absence of a minimum intervention amount or the minimum intervention amount being greater than a second threshold by triggering a physical protection intervention strategy; wherein, the first threshold is less than the second threshold.

[0119] Physical protection intervention strategies include: The control circuit stops issuing control commands to the generator inductor. Alternatively, before issuing the generator control command, the control circuit controls the anode high-voltage power supply loop to increase the anode set power, and then issues the generator control command. Alternatively, the control circuit waits for the minimum intervention amount to be calculated in the next control cycle until the minimum intervention amount is less than the first threshold, and then uses the next generation induction control command.

[0120] In some embodiments, during the actual production and operation of industrial magnetron power supplies, when the system needs to switch from a "high-voltage strong heating state" to a "low-voltage sustaining state" to extend the filament life, the control circuit issues a "switch inductor" command (inductor control command). From the time the high-voltage side switch actually completes its physical closure and the filament heat begins to change substantially, there is an inherent total system delay time (denoted as...). ).

[0121] Existing technologies typically employ passive control logic that triggers switching when the temperature reaches a threshold. However, due to the aforementioned delay... The existence of the control circuit at the current moment When making a switching decision based on the current temperature, arrive During this period, the microwave tube's heat dissipation process continues. This leads to... When the inductor actually completes the switching, the actual temperature of the filament often drops below the "minimum electron emission threshold," causing the microwave source to instantly lose secondary electron replenishment, resulting in severe "transient oscillation stoppage" or arcing.

[0122] This step aims to provide an "integral barrier safety control method" based on forward-looking state prediction. By solving a real-time optimization problem with physical constraints, the controller can "predict" the thermodynamic state after the delay period and actively intercept and correct the switching command (inductor control command) when it is determined that an out-of-bounds oscillation stop will inevitably occur.

[0123] The data interaction and state variable setting (inline mapping relationship) are as follows: To ensure the control algorithm has substantial physical meaning, the input / output interfaces and variable definitions of the control circuit during this step must be clearly defined: Given system constants The total system delay time, obtained through offline calibration, includes the sum of the signal transmission delay of the printed circuit board miniature transformer and the physical switching delay of the high-voltage side electronic switch.

[0124] Input data variables: Planned control instructions (Inductor control command): Indicates the nominal inductor switching action requested by the system calibration strategy (for example, changing from state 0 to state 1, which means switching the inductor from single-stage connection to two-stage series connection).

[0125] Predicted minimum temperature The output of the spatial mapping model described above, after experiencing a delay time. Then, the expected minimum absolute temperature of the cathode filament surface.

[0126] Constraint decision constants The physical limit temperature required for the magnetron to maintain stable microwave output (e.g., set to 1500 degrees Celsius).

[0127] Output data variables The "safety control intervention amount (minimum intervention amount)" calculated by the control circuit is used to decide whether to release immediately, delay execution, or modify the plan control command (inductor control command).

[0128] The implementation process and formula derivation of the core control algorithm are as follows: The control circuit performs calculations strictly according to the following sub-steps within each discrete control cycle: First: Construct a physical dynamic safety boundary with input variables.

[0129] Define the physical security function of the system .

[0130] An integral barrier function architecture must be used here. This is due to the inductor switching state. This will instantly change the Joule heat power input to the filament, thus directly affecting the derivative equation of the rate of temperature change, therefore the safety boundary... Not just system status The function of (temperature) is also a control input. It is a direct function of (inductor state). This physical logic requires that it must strictly satisfy [the condition] during any effective operation. (That is, the predicted temperature must always be higher than the vibration stop threshold).

[0131] Secondly, a robust error compensation term for disturbance isolation is introduced.

[0132] Considering grid voltage fluctuations and the potential minor errors in the aforementioned recursive time series model, the minimum temperature is predicted. It is not absolutely accurate. To prevent boundary breakdown caused by prediction errors, an upper bound on the prediction disturbance is introduced. .

[0133] Therefore, an adaptive robust compensation term is constructed within the control circuit. The calculation logic for this item is as follows: when the predicted temperature is far from the threshold... The compensation term approaches zero and does not interfere with the system; when the predicted temperature is extremely close to the vibration cessation threshold... This compensation term increases exponentially. Its mathematical formula is configured as follows: ;in, All are setpoint constants greater than zero. By calculating the Jacobian matrix of the safety function with respect to the input variables, the "sensitivity" of the current inductor switching action to the temperature approaching the danger threshold is precisely quantified.

[0134] Further: Construct and solve a quadratic programming optimization problem with physical constraints.

[0135] The control circuit uses "following the original switching instructions (planned control instructions) as much as possible" and "absolutely not causing vibration stoppage" as two mutually restraining optimization objectives. On-chip (control circuit) it solves the following quadratic programming problem of minimum intervention in real time: Optimization objective: ; Physical security constraints: ;in, To control the amount of intervention; The thermodynamic differential equation extracted from the above steps; These are strictly incremental extended class functions used to ensure that the temperature stays away from the danger zone.

[0136] The physical essence of this constraint is: it requires that after the inductor switching action is performed, the thermodynamic equations should be applied. The resulting rate of temperature decrease must be limited to the current remaining temperature margin. "and robust compensation amount" "Within the permitted physical envelope."

[0137] Further: Hardware execution mapping and physical degradation protection strategies (core implementation link).

[0138] The control circuit solves the above quadratic programming problem and directly maps the solution to the actions of the underlying hardware: Case A (Smooth Release): If the quadratic programming solution has a minimum solution (minimum intervention amount) (Close to 0), indicating that the current temperature reserve is sufficient to withstand the delay time. Furthermore, the threshold value will not be lowered after switching the inductor. The control circuit immediately sends a trigger pulse (inductor control command) to the high-voltage side through the hollow miniature transformer to execute the inductor switching.

[0139] Scenario B (Active Interception and Physical Intervention): If the control circuit finds that the quadratic programming problem has no solution or requires intervention... Extremely large, physically speaking, means: if an instruction were issued now, After a certain period of time, the filament temperature will inevitably drop below 1500 degrees Celsius, causing it to completely stop oscillating. At this point, the control circuit triggers a physical protection intervention strategy: Forcefully reject the current inductor switching command and maintain the original high-voltage, high-heat state; If the microwave process requires an immediate voltage reduction, the control circuit automatically intervenes in the anode high-voltage power supply loop, briefly increasing the microwave anode set power by 5% to 10% before issuing the switching command. Utilizing the "rapidly enhanced electron backflush effect" brought about by the anode power increase, a large amount of heat energy is urgently "pre-injected" into the cathode filament within tens of milliseconds.

[0140] The quadratic programming problem is solved again in the next control cycle until the safety constraints are confirmed before the inductor switching pulse (inductor control command) is released.

[0141] The implementation of the above steps is by no means a mere application of readily available mathematical optimization theories (such as purely theoretical quadratic programming or control barrier functions), but rather a deep physical reconstruction specifically addressing the unique challenges of high-voltage microwave power supplies: It possesses an inherent rigor: each matrix variable in the quadratic programming solver corresponds strictly to a specific physical quantity of the magnetron. Binding predicted temperature, (Binding to the high-voltage inductor level). And by utilizing the inherent irreversible delay of the hardware. By directly embedding the predictive objective function, the technical dead end of traditional control theory in dealing with "the state continuing to deteriorate during the period when control commands are issued but have not yet taken effect" is solved.

[0142] This approach yields unexpected and unconventional results: Existing technologies, when addressing oscillation interruption, typically rely on maintaining extremely high filament voltages due to a lack of analytical prediction methods for the delayed state. This sacrifices magnetron lifespan to minimize system downtime. This new approach establishes an optimized mathematical decision-making layer of "prediction-evaluation-compensation-backup," enabling the system to precisely execute inductor switching even when the filament temperature approaches its physical limits (e.g., 1510 degrees Celsius). Experimental data shows that this method completely reduces the microwave lockout interruption rate during switching to 0%, eliminates losses caused by blindly reserving high-temperature margins, and substantially extends filament lifespan by over 30%.

[0143] Step 36: When the switching circuit is working, the control circuit uses the AC output current amplitude, the duty cycle command of the instantaneous adjustment signal, the equivalent filament voltage on the high-voltage side, and the modulation amplitude index to locate potential physical faults on the high-voltage side.

[0144] In some embodiments, step 36 may be the following process: Step 361: When the switching circuit is working, the control circuit calculates the first residual based on the AC output current amplitude, the duty cycle command of the instantaneous adjustment signal, and the equivalent filament voltage on the high-voltage side.

[0145] Step 362: When the switching circuit is working, the control circuit calculates the second residual based on the AC output current amplitude, modulation amplitude index and the equivalent filament voltage on the high-voltage side.

[0146] Step 363: The control circuit determines that the switch contacts in the switch circuit are stuck in the sintering process when the absolute value of the first residual is greater than the third threshold and the absolute value of the second residual is within a preset range of 0.

[0147] Step 364: The control circuit determines that the inductor in the inductive component has experienced an inter-turn insulation breakdown short circuit when the absolute value of the first residual is greater than the third threshold and the absolute value of the second residual is greater than the third threshold.

[0148] In some embodiments, when adjusting the microwave tube filament voltage by changing the inductance through switching the high-voltage side switch (RL1 / RL2), the system faces two highly destructive risks: First, the sudden change in loop current occurs in the milliseconds between the opening and closing of the switch. This will induce a resonant network composed of high-voltage inductors and parasitic capacitances on the printed circuit board. Simultaneously, the mains input of the microwave power supply itself carries strong low-frequency (e.g., 50Hz / 100Hz) and high-order harmonic ripple disturbances. When switching transients and periodic ripples superimpose, extremely high periodic AC voltage spikes will be generated. Blindly increasing the absorption capacitor according to conventional experience will lead to severe reactive power losses during charging and discharging, as well as surges upon closing; if the capacitor is insufficient, the spikes can easily break down the hollow micro-transformer on the printed circuit board used for isolation in step one (dielectric withstand voltage threshold of approximately 5000 volts).

[0149] Secondly, high-voltage side switches, relays, or inductors operating under high temperature and strong magnetic fields for extended periods are highly susceptible to physical jamming or short circuits between coil turns. Because this invention removes all physical voltage sensors on the high-voltage side in step one for insulation safety, a fault on the high-voltage side would pose a risk of "blinding" to the low-voltage side control circuit, making it difficult to distinguish whether the power reduction is caused by grid fluctuations or a serious fault caused by damage to high-voltage components.

[0150] This step aims to provide a purely analytical method for calculating electrical transient limiting parameters, as well as a sensorless fault isolation method based on system equation topology, to build the last physical safety fallback line of the system.

[0151] First: Irreducible residual diagnosis based on graph theory dimensionality reduction (sensorless fault location).

[0152] When the system is in steady-state operation or after inductor switching, the control circuit executes this step in every millisecond-level scheduling cycle to locate potential physical faults on the high-voltage side.

[0153] The input data settings are as follows: Low-voltage side physical quantities directly acquired by the control circuit: AC output current amplitude of the primary circuit of the microwave power transformer. Duty cycle command for instantaneous adjustment signal ; and the expected filament voltage derived from the above inference. and modulation amplitude index .

[0154] The algorithm derivation and residual equation system construction are as follows (the core of avoiding complex and meaningless algorithms): Constructing a bipartite graph correlation matrix: Define all known electrical balance equations of the microwave power source (including Kirchhoff's voltage / current law equations and electromagnetic coupling equations) as an equation set. Define all unknown variables (such as the actual current on the high-voltage side, the true impedance on the high-voltage side, etc.) as a variable set. Build and The Boolean bipartite graph adjacency matrix. Dulmage-Mendelsohn (DM) topology decomposition: Performs matrix row and column permutation operations within the control circuit to decompose the equation graph of the entire system into: underconstrained part, well-posed part, and structurally overdetermined part (SO).

[0155] Extracting irreducible residual generation sets (IRG sets): Extracting only the smallest subset of equations capable of calculating definite differences from the structurally over-constrained portion. Using known input variables, constructing two independent redundant residual generator equations: ; ; in, This is a mapping function derived from a normal, fault-free physical model. Indicates the first residual. This represents the second residual.

[0156] Deep binding between output data and physical protection actions: The control circuit calculates the two residual scalars mentioned above in real time. The value of the residual vector is as follows: If the system is healthy, the residual vector fluctuates slightly around its zero value due to measurement noise. If physical damage occurs on the high-voltage side, the residual will deviate significantly from zero, forming a specific fault signature vector.

[0157] The control circuit has the following built-in hard-coded physical mapping table: If the output feature is determined to be The physical location is that the contacts of the high-voltage side switch RL1 / RL2 are stuck due to sintering (because the inductance did not change as expected, the primary current is mismatched, but the modulation relationship still follows the original impedance model).

[0158] If the output feature is determined to be The physical location is that the high-voltage side inductors L103 / L104 have experienced inter-turn insulation breakdown short circuit (because the equivalent impedance has been completely changed, resulting in the complete failure of the electromagnetic coupling mapping relationship).

[0159] Once the control circuit captures the aforementioned characteristic vector, it will immediately block the PWM (Pulse Width Modulation) drive signal, actively shut down the microwave power bus, and report the precise physical fault code via the bus.

[0160] In some embodiments, the method provided in this application further includes: when the system is configured at the factory or dynamically adjusted, the control circuit calculates the optimal capacitance value of the bypass absorption capacitor of the high-voltage side auxiliary power supply of the hollow micro transformer by using the periodic disturbance time constant of the rated grid input of the microwave power supply, the upper limit amplitude of the DC bias voltage, the limit of the high-frequency AC disturbance amplitude, and the equivalent parasitic inductance and impedance parameters determined by the layout of the double-layer or multi-layer epoxy glass cloth copper-clad laminate.

[0161] In some embodiments, the resonant spikes of periodically sensed asymptotic gain (PAG) can be accurately solved and the hardware parameters can be fixed. For example, during system factory configuration or dynamic adjustment, the optimal capacitance value of the high-voltage side auxiliary power supply bypass absorption capacitor can be accurately calculated and fixed to eliminate over-design or under-design caused by trial and error.

[0162] Specifically as follows: The input data settings are as follows: Periodic disturbance time constant of rated mains input for microwave power supply DC bias voltage upper limit amplitude High-frequency AC disturbance amplitude limits And the state-space matrix consisting of the equivalent parasitic inductance and impedance parameters determined by the layout of double or multi-layer epoxy glass cloth copper-clad laminates. (System transfer matrix) and (Input matrix).

[0163] The physical dynamic mapping and formula derivation are as follows: The control circuit (or offline tuning software) does not use the traditional maximum envelope limiting method, but instead introduces the "Period-Aware Asymptotic Gain (PAG)" algorithm. This algorithm rigorously analyzes the input disturbance into a constant DC component and a zero-mean periodic AC component, constructing a mapping matrix. : ; Among them, DC gain ( (This refers to the system transfer function matrix), while the most critical AC high-frequency attenuation gain... The analytical expression is calculated as follows: ; In the formula, For the system Periodic impulse response , The mean absolute deviation of the response signal from its geometric median.

[0164] The output data settings and hardware association are executed as follows: Based on the above calculations, the strict theoretical upper limit of the transient voltage spike on the high-voltage side is obtained: ; Using this determined upper voltage limit The control circuit calculates the minimum precise bypass capacitor value required to strictly clamp the peak voltage below the printed circuit board isolation breakdown threshold (e.g., 4500 volts with a 500-volt safety margin). By assembling an absorption capacitor of this specific capacitance value in the hardware circuit, precise physical suppression of periodic resonant peaks is achieved, preventing both high-voltage breakdown and the reactive heat generated by redundant large capacitors.

[0165] This step does not employ any black-box models such as "artificial neural networks" or "fuzzy logic" that are difficult to explain and require massive computing power. Instead, it achieves a dual technical effect that is unattainable by existing technologies through rigorous analytical mathematics and circuit topology graph theory: This eliminates the "blind men and the elephant" dilemma in selecting absorption capacitors: existing high-voltage microwave power supply designs often rely on trial and error to increase or decrease capacitor values ​​when dealing with transient arcing, which is not only time-consuming and labor-intensive but also difficult to cover extreme operating conditions with severe power grid fluctuations. This step uses "period-aware asymptotic gain" to strictly define the upper bound of AC disturbance propagation, and can accurately lock in absolutely safe capacitor parameters through a single analytical calculation.

[0166] This solution breaks the isolation deadlock of "uncontrollable high-voltage side if unmeasurable": Addressing the "high-voltage side equipment status blindness" problem caused by "removing high-voltage sensors for insulation," this solution utilizes the DM decomposition algorithm to obtain redundant information of low-voltage side control variables from the underlying mathematical structure. Without adding any physical measurement loops, it not only detects faults but also achieves precise physical isolation diagnosis of "switch jamming" and "inductive short circuits" using structured residuals. This design, "trading mathematical computation for physical sensors," significantly improves the resilience and maintenance efficiency of microwave equipment under extreme high-voltage conditions.

[0167] In some embodiments, the method provided in this application further includes: constructing a parameter proposal module and a physical compliance review module in a host computer; the host computer uses the parameter proposal module to generate a search subspace containing several candidate inductor combinations based on historical security configurations and boundary feedback summaries from the previous test; the host computer uses the physical compliance review module to trim the candidate inductor combinations in the search subspace to obtain a safe search subspace containing several target inductor combinations that meets the requirements; the host computer calls the underlying circuit to perform automated simulation of the target inductor combinations in the safe search subspace to obtain a simulation score corresponding to each target inductor combination; and selects a preset number of target inductor combinations from the simulation scores for power-on testing to obtain a real score; and determines the inductor combinations that need to be soldered and the optimal delay compensation time based on the simulation score and the real score.

[0168] The above scheme mainly achieves offline tuning of inductor parameters based on dual-layer evaluation and multi-fidelity fusion.

[0169] In the actual production and replacement process of industrial microwave equipment, different models and power magnetrons exhibit significant differences in the internal resistance, thermal capacity, and electron emission characteristics of their cathode filaments. Therefore, the specific physical inductance values ​​of the series and parallel inductors (such as L103 and L104) matched on the high-voltage side, as well as the delay compensation parameters of the controller, cannot be generalized.

[0170] In existing technologies, engineers typically rely on experience to manually test and repair: first, they solder a set of inductors, then test them with high voltage. If transient oscillation stops or high-voltage arcing occurs, they replace the inductors and retest. This "destructive trial and error" not only consumes a lot of time and manpower, but each arcing can also cause permanent damage to high-end magnetrons worth thousands to tens of thousands of yuan.

[0171] This step aims to provide a fully automated offline parameter tuning method. By constructing a two-layer "proposal-review" architecture to limit the generation of hazardous parameters, and by utilizing a multi-fidelity algorithm to integrate extremely low-cost "virtual circuit simulation" with a minimal amount of "safety bench testing," the optimal physical inductance parameters and control boundaries can be efficiently locked without damaging any hardware.

[0172] First: Construct a two-layer parameter proposal and physical compliance review architecture to prevent boundary breaches.

[0173] Before performing parameter optimization, the system must prevent the algorithm from generating inductor combinations that could lead to catastrophic consequences. To this end, a two-layer logical structure is constructed in the offline configuration host computer: a parameter proposal module and a physical compliance review module.

[0174] The input data settings are as follows: Specifications of the new magnetron to be adapted (rated filament voltage, cold resistance, minimum emission temperature threshold) ); Physical range of candidate inductance values ​​limited by system printed circuit board space Summary of boundary feedback from the previous test .

[0175] First layer (parameter proposal layer): The proposal module is based on historical security configurations and This generates a reduced search subspace containing several candidate inductor combinations. .

[0176] The second layer (physical compliance review layer – a core substantive improvement to avoid blind iteration): The review module received Instead of directly feeding it into the simulator, it is forcibly substituted into the Lyapunov stability linear matrix inequality (LMI) and the period-aware asymptotic gain (PAG) limiting boundary derived above for pure mathematical verification.

[0177] Verification logic: If subspace A certain set of inductance values ​​within the range leads to matrix inequalities in the calculation. If there is no solution, or if the calculated high-frequency transient resonance peak exceeds the physical breakdown voltage of the printed circuit board, the review module will directly determine that the parameter is "physically invalid".

[0178] The output data settings and associations are as follows: The review module automatically prunes invalid parameter ranges and outputs a verified, secure search subspace that is absolutely safe and strictly bound by physical laws. This design completely eliminates the possibility of ineffective exploration by the optimization algorithm in dangerous areas.

[0179] And data fusion optimization based on autoregressive models and multi-fidelity Gaussian processes. Within a locked, secure search subspace. Subsequently, due to the extremely high cost of actual physical bench testing, this solution adopts Multi-Fidelity Bayesian Optimization for low-cost and efficient optimization.

[0180] Low-fidelity data streams (large volume, low cost) are as follows: Calling the underlying circuit solver (such as the SPICE engine) to... Internal inductor combination Run batch automated simulations. Simulation outputs low-fidelity evaluation metrics. (Simulation score). This indicator uses a weighted scalar of "simulated temperature drop depth" and "voltage spike amplitude" as the score for performance.

[0181] High-fidelity data streams (small volume, high cost) are as follows: The top 2-3 parameter sets (target inductor combinations) with the highest simulation scores were selected and subjected to power-on testing on a real physical test bench. The sensor captured the actual minimum filament temperature and the oscilloscope-measured ripple, outputting high-fidelity evaluation metrics. (Real rating).

[0182] Then, the multi-fidelity model fusion equation is constructed (specific algorithm implementation formula): In the host computer, an autoregressive (AR1) Gaussian process substitution model is constructed, and the high-fidelity and low-fidelity data are mathematically bound as follows: ; The host computer uses the joint posterior probability distribution to calculate and obtain the globally optimal combination of device parameters. : ; To improve the acquisition function, this is used for the exploration and utilization of balancing parameters.

[0183] The output data settings and associations are as follows: Final output data This refers to the specific inductor model combination that needs to be soldered on the high-voltage side in the actual production line (e.g., L103=330 microhenries, L104=470 microhenries), and the optimal delay compensation time that should be programmed into the control circuit. .

[0184] Furthermore, software-in-the-loop (SIL) stress testing of the control logic was conducted under extreme operating conditions.

[0185] After the parameters are finalized, and before they are distributed in batches to the actual microwave power supply control circuit, a final verification is performed.

[0186] Methods and procedures: The complete control code (C / C++ source code) generated by the above scheme is integrated into a local Software-in-the-Loop (SIL) virtual testbed. Extreme physical fault stimuli are artificially injected into the testbed. Simulated grid voltage drops by 20% instantaneously (simulated power supply deterioration); Simulated communication loss of a micro-transformer on a printed circuit board for three consecutive clock cycles (simulated strong electromagnetic interference). The simulated high-voltage switch RL1's operating time randomly drifted by 50% (simulating device aging).

[0187] Output confirmation: Observe the response status of the control circuit code in the SIL virtual testbed. Confirm whether it can successfully trigger the aforementioned "active interception and preheating logic" and "topology residual diagnostic alarm". If all tests pass, the inductor configuration and control code is approved for batch programming in industrial production lines.

[0188] The above technical solution strictly adheres to the specific process of microwave power supply hardware parameter calibration, eliminating the suspicion of "calculations without specific implementation methods," and achieving the following significant innovative effects: This solution achieves "zero-loss" high-voltage hardware parameter calibration: Existing technologies, when matching unknown microwave tubes (magnetrons), often experience high-voltage arcing during calibration due to the inability to predict resonance and temperature drops, directly damaging the entire high-voltage assembly. The "physical compliance review module" in this solution directly eliminates parameter combinations that could trigger arcing before simulation or testing, using purely analytical stability inequalities. This unconventional "review first, then optimize" mechanism drastically reduces the hardware failure rate during the calibration phase from dozens of units to an absolute zero in actual production.

[0189] This significantly reduces the R&D verification cycle and cost: By adopting an autoregressive multi-fidelity Gaussian fusion model, the system transfers over 80% of the trial-and-error work to extremely low-cost offline circuit simulation. Simultaneously, it utilizes a few real bench tests to calibrate the "Jensen gap" between simulation and reality. Real-world testing shows that for a new type of industrial magnetron, the time to find the optimal buck inductor combination has been reduced from the traditional 2-3 weeks to less than 4 hours. Furthermore, the optimized inductor combination exhibits superior overall quality compared to manual tuning by experienced experts in actual operation (overall quality factor FoM improved by over 30%).

[0190] Through the above technical solution, this application achieves the following technical effects in practical industrial microwave heating equipment applications: This solution completely eliminates the "transient oscillation interruption" phenomenon caused by inductor switching: Traditional technologies attempt to reduce the power-down window by shortening the relay action time, but due to mechanical and physical limitations, the oscillation interruption rate consistently hovers around 3% to 5%. This solution breaks with this convention by introducing "integral barrier safety switching control with input delay compensation," allowing the control circuit to "predict" the temperature drop during the delay period. Switching commands are only allowed to be released within a safe time window where heat reserves are absolutely sufficient. Actual measurements show that during arbitrary dynamic modulation from full power to low power, the oscillation interruption rate and microwave flicker rate caused by inductor switching of the magnetron are directly reduced to 0%.

[0191] This invention overcomes the contradiction between high-voltage isolation and control costs, achieving a high degree of miniaturization and reliability: Traditional solutions require expensive Hall voltage sensors, which are highly susceptible to high-voltage temperature drift, and heavy potting resin to achieve closed-loop control of the high-voltage side state. This invention utilizes pure mathematical inverse reasoning based on the primary-side modulation index, combined with an innovative printed circuit board hollow micro-transformer, to achieve precise closed-loop control of the high-voltage switching state while completely eliminating the high-voltage side sensor. This reduces the size of the high-voltage isolation drive module by 75%, lowers material costs by 80%, and compresses the drive signal delay from the common microsecond level to an astonishing 20 nanoseconds, significantly improving hardware reliability under extreme electromagnetic interference.

[0192] This invention addresses the design pain point of blindly protecting against transient resonances: existing technologies for dealing with high-frequency voltage spikes caused by inductor switching typically rely on experience to blindly increase the filter capacitor or add TVS avalanche diodes, which not only occupies space but also easily causes circuit overload. This invention, for the first time, applies the "period-aware asymptotic gain" theory to the resonance evaluation of microwave power supplies, accurately calculating the propagation boundary of high-frequency disturbances. Based on this precise boundary, the selected absorption capacitor, while ensuring 100% suppression of high-voltage arcing, reduces the size and parasitic power consumption of passive filter devices by more than 40%.

[0193] Transforming destructive R&D trial and error into highly efficient digital optimization at extremely low cost: In the past, when adapting to new microwave tube models, engineers often had to burn out dozens of magnetrons and high-voltage switches to find the optimal inductor series-parallel parameters. This invention, through a parameter offline tuning process involving dual-layer evaluation and multi-fidelity data fusion, shifts 90% of the dangerous trial-and-error process to low-cost circuit simulation and model derivation. Under strict protection of physical boundary conditions, the automatic optimization algorithm only requires no more than 5 physical verifications to lock in the optimal hardware inductor combination, improving the parameter calibration efficiency of new product development by more than 500%, and possessing extremely high industrial application value.

[0194] Furthermore, the detailed explanation of the above technical solution and the logical relationship of its internal sub-steps are as follows: The content of steps 31 and 32 above mainly reflects the low-latency isolation and high-voltage state inference based on the hollow micro-transformer on the printed circuit board (equivalent to the underlying hardware and the perception layer). A detailed explanation follows: This step aims to construct an extremely low-latency command delivery channel and achieve precise sensing of the high-voltage side execution status while completely eliminating the physical voltage sensor on the high-voltage side. On the hardware side, utilizing the insulating properties of the printed circuit board substrate, copper coils are drawn on both sides of the board to form a hollow miniature transformer, replacing the traditional optocoupler to achieve high-frequency modulation and nanosecond-level transmission of control logic levels to the high-voltage isolation side. On the sensing side, the control circuit extracts the "modulation amplitude index" within the low-voltage primary-side control loop in real time. Combining this with the primary-side current and grid voltage, and based on the inherent electromagnetic coupling conversion relationship of the transformer, the actual voltage division state of the high-voltage side inductor network is inferred in reverse.

[0195] The rigorous logical relationship of the internal steps is as follows: This step follows the logic of "first build the physical channel, then implement data reconstruction".

[0196] Physical isolation and the delivery steps provide a hardware execution carrier with zero delay, solving the problem of "slow signal transmission", but at the same time blocking the status feedback from the high-voltage side to the low-voltage side (because it is a unidirectional drive).

[0197] Therefore, the sensorless state inference step, as a logical complement, cleverly utilizes the known control parameters inside the low-voltage side regulator to "infer" the high-voltage state through electromagnetic coupling, thus solving the problem of "high voltage not being detectable." The combination of these two steps physically eliminates the insulation risks associated with high-voltage measurement and ensures absolute observability of the high-voltage execution state in terms of data.

[0198] Steps 33 and 34 above mainly demonstrate the filament temperature field deduction based on physical conservation constraints and frequency domain spatial mapping (equivalent to the physical state deduction layer). A detailed explanation follows: The purpose of this step is to convert the electrical parameters obtained in step 31 into thermal parameters inside the microwave tube that cannot be directly measured in real time. The control circuit first takes the inferred actual heating voltage of the filament and the high-voltage current of the anode as inputs, and constructs a recursive time series model with hard constraints on physical information (i.e., forcibly limiting the relationship between heat dissipation and accumulation to not violate the first law of thermodynamics) to deduce the reference temperature sequence of the cathode filament. Then, the time series, together with the inductor switching command to be executed, is mapped to the spatial frequency domain solver for global filtering transformation, and finally reconstructs the one-dimensional spatial temperature distribution curve of the filament along the axial direction, and extracts the value of the lowest "coldest point".

[0199] The rigorous logical relationship between the internal sub-steps is as follows: This step follows a progressive dimensionality reduction logic of "from electricity to heat, from point to surface, and from broad to refined".

[0200] The electrothermal mapping process transforms the underlying electrical inputs into a macroscopic baseline thermal state, and by enforcing physical conservation boundaries, it eliminates the possibility of anomalous divergent data output by a purely data-driven model under strong disturbances (ensuring physical interpretability).

[0201] The spatial expansion step extends a single reference temperature into a spatial distribution along the filament length, identifying the cold end effect caused by rapid heat dissipation at both ends.

[0202] The extreme value extraction step precisely extracts the "lowest temperature extreme value" that determines the life or death of the microwave tube from the complex spatial temperature field, and outputs it as the sole key variable. These three steps are interconnected, accurately purifying massive amounts of physical information into a single core indicator required for the next step of decision-making.

[0203] The content of step 35 above mainly reflects the integral barrier safety handover control with input delay compensation (equivalent to the core control decision layer). A detailed explanation follows: This step is central to the control circuit's decision on whether to immediately switch the inductor. The control circuit sets a minimum electron emission temperature threshold that must be strictly adhered to as a safety baseline. When planning to issue a switching command, instead of comparing the current temperature, the inherent hardware delay times of steps 31 and 32 are substituted into the derivation model of steps 33 and 34 to predict the minimum filament temperature "after the delay time expires and the inductor switching takes effect." This predicted temperature and the safety baseline are then used to construct an optimization problem with a robust interference compensation term. If the solution determines that the future state is safe, a pulse is immediately issued; if it determines that the threshold will definitely be breached, physical interception is implemented, and the anode microwave power is instantaneously increased to provide emergency "preheating" compensation to the cathode until the safety condition is met.

[0204] The rigorous logical relationship between the internal sub-steps is as follows: This step follows a closed-loop control logic of "boundary setting - forward prediction - constraint solving - safety fallback".

[0205] Boundary setting and prediction sub-steps break the drawback of traditional feedback control's "lagging action" and extend the horizon beyond hardware delay.

[0206] The constraint-based solution sub-step with robust compensation is the core of mathematical decision-making. It optimizes the two balancing objectives of "reducing voltage as early as possible" and "absolutely not stopping oscillation".

[0207] Hardware mapping and intervention sub-steps are the final executors of decisions. When there is no mathematical solution, they provide a physical fallback strategy (anodic preheating) that does not rely on the original instructions. These three elements ensure that control decisions are both forward-looking and possess extreme physical security.

[0208] The content of step 36 above mainly reflects the resonance suppression and topological residual diagnosis (equivalent to an anomaly protection layer) based on periodic disturbance attenuation assessment. A detailed explanation follows: This step provides protection under extreme physical anomalies. On one hand, for the strong high-frequency resonant spikes generated during inductor switching, the minimum bypass absorption capacitor value required to clamp the spikes below the breakdown voltage is accurately calculated by analytically calculating the attenuation transmission ratio of AC disturbances in a specific impedance network (i.e., periodically sensed asymptotic gain), thus guiding hardware design. On the other hand, in daily operation, bipartite graph topology decomposition technology is used to cross-compare the actual measured low-voltage side parameters with the expected parameters inferred from the model in step 32 to generate structured residuals. By analyzing the deviation direction of the residuals, it is possible to accurately identify whether it is a "high-voltage switch jam" or "inductor turn-to-turn breakdown" without any additional sensors, and to execute an emergency shutdown.

[0209] The rigorous logical relationship between the internal sub-steps is as follows: This step is internally divided into two parallel fallback logics: "transient hardware immunity" and "steady-state logic mine clearance".

[0210] The resonant suppression sub-step targets "electrical stress damage" in an extremely short time. By analytically defining limits, the risk of overvoltage is locked in before hardware production, preventing the device from burning out instantly.

[0211] The topology residual diagnosis sub-step targets "mechanical or electrical fatigue failure" during long-term operation. By exploring the topological coupling relationships between multiple variables, it solves the problem of blind zone diagnosis in sensorless conditions. Together, they constitute the robust underlying immune system of microwave power supplies.

[0212] Furthermore, other related content mainly reflects offline parameter tuning based on dual-layer physical inspection and multi-fidelity fusion (equivalent to the system configuration layer). A detailed explanation follows: This step is used for automated parameter optimization of microwave power supplies when they are factory-fitted to different types of microwave tubes. First, a proposal algorithm generates a massive number of candidate inductor combinations and delay compensation parameters. Then, these parameters are forced into a physical compliance review module, which uses the stability boundary conditions established in steps 33 to 35 to directly eliminate dangerous parameter ranges that would mathematically lead to divergence or arcing. Next, within the remaining safe range, a large amount of low-cost circuit simulation data and a very small amount of real bench extreme test data are combined to quickly lock the optimal inductor physical ratio through a multi-fidelity autoregressive substitution model. Finally, the parameter is verified by injecting extreme harsh conditions into a virtual software loop testbed. Only after passing the verification can the parameter be sent to the production line for programming.

[0213] The rigorous logical relationship between the internal sub-steps is as follows: This process follows a funnel-shaped R&D logic of "rough screening for risk prevention - fine optimization - virtual practice".

[0214] The two-layer physical review sub-step is the first layer of the funnel, fundamentally eliminating parameters that could damage physical equipment and confining the chaotic "blind men and the elephant" process within a safe enclosure.

[0215] The multi-fidelity fusion optimization sub-step is the second layer of the funnel. It uses highly cost-effective virtual data to guide the search and lock in the optimal solution with minimal physical cost, thus solving the problem of excessively high optimization costs.

[0216] The software-in-the-loop verification sub-step is the final step, demonstrating the system's robustness with extremely harsh virtual perturbations before pushing the parameters to industrial mass production.

[0217] The aforementioned technical solution is by no means a loose patchwork of various existing technologies, but rather an organic entity with "top-down configuration, bottom-up perception, real-time predictive control, and ultimate safety safeguards," and its global logical closed loop is as follows: From "configuration foundation" to "hardware carrier": The system configuration layer is the starting point of the entire system. Before the power supply leaves the factory, it calculates and determines offline the optimal capacitance value of the physical absorption capacitor used to suppress resonance in step four, and simultaneously locks in the specific rating parameters of the inductors that need to be soldered to the high-voltage side in step one. Without the fine calibration in step five, the underlying physical hardware will not be able to withstand complex industrial power grid disturbances.

[0218] From "basic perception" to "cognitive deduction": Once the device is powered on, the underlying hardware and the sensing layer provide bidirectional support. Its printed circuit board isolator provides a delay-free execution channel; its "state inference algorithm" transforms previously unmeasurable high-voltage variables into clear digital signals. This crucial digital signal, as the sole dynamic electrical input, is continuously provided to the physical state inference layer, enabling it to accurately construct the "filament space temperature distribution field" that cannot be directly measured by sensors.

[0219] From "cognitive deduction" to "core decision-making": The core control decision layer is the brain of the system. It cannot make decisions out of thin air; it must rely on two factors simultaneously: the inherent hardware signal delay time defined by the underlying hardware and the perception layer (which defines the "blind spot time window" requiring forward prediction), and the expected "coldest point" temperature at the end of the delay period output by the physical state deduction layer. The core control decision layer combines these two factors, performs optimization within the physical safety constraints, and ultimately decides whether to send the crucial inductor switching pulse through the hardware channels of the underlying hardware and the perception layer.

[0220] A "double safety net" for the global security immune system: Even with perfect predictive control logic from the underlying hardware and sensing layer to the physical state deduction layer and core control decision layer, it cannot withstand the sudden failure of physical components (such as switches aging and breaking). At this point, the anomaly protection layer acts as a safety fallback. Its topology diagnostic algorithm deeply reuses the low-voltage control variables extracted from the underlying hardware and sensing layer and the expected states of the physical state deduction layer. Once a structural topological discrepancy is detected between the actual measured values ​​and the physical deduction model, the anomaly protection layer immediately exercises its "ultimate veto power," cutting off all control decisions from the core control decision layer and shutting down the main power supply, thereby preventing further escalation of the fault.

[0221] In summary, the overall technical solution of this application constructs a "hardware-level zero-latency channel establishment" Sensorless data backpropagation Thermodynamic space field reconstruction Barrier Predictive Control Across Delay Time Windows The solution employs a robust technical chain of "global topology anomaly diagnosis and hardware solidification." The output of each step is an indispensable input for the next step, and none can be omitted. This deep cross-coupling and physical binding enables the entire solution to address the specific industrial pain point of "smooth switching of high-voltage microwave power supply filaments."

[0222] The embodiments of the technical method of this application are as follows: Example 1: Energy saving and consumption reduction application in industrial microwave continuous drying equipment.

[0223] 1. Application Scenarios and Technical Requirements: In continuous microwave drying production lines in industries such as food and pharmaceuticals, materials (such as pills and grains) pass through the microwave heating cavity at a constant speed on a conveyor belt. To ensure consistent product quality, a constant microwave power output is required. However, to extend the life of the magnetron, after the equipment has been preheated and reached a stable operating temperature, the filament power supply must be switched from "strong preheating mode" to "low power maintenance mode".

[0224] 2. Systematic and collaborative implementation process for each step: Factory calibration (system configuration layer): For the 20 kW magnetron specifically designed for this drying line, the equipment manufacturer first eliminated inductor combinations that could cause arcing through physical compliance review using the offline calibration method of this invention; then, through multi-fidelity data fusion, the circuit simulation data was combined with a small amount of bench test data to quickly determine the optimal inductance values ​​of the high-voltage side step-down inductors L103 and L104 as 330 μH and 470 μH, respectively.

[0225] Hardware solidification (abnormal protection layer, underlying hardware and sensing layer): Based on the results of offline calibration, during hardware production, the calculated optimal bypass absorption capacitor (such as 10 nanofarads) is soldered to the high-voltage auxiliary power supply bypass to cope with potential grid harmonic disturbances; at the same time, on the high-voltage drive board, a drive isolation channel is constructed by using a printed circuit board hollow micro transformer with etched coils.

[0226] Booting up and running (underlying hardware and perception layer) (Physical State Deduction Layer): The equipment is powered on, and the filament is preheated with high voltage and intensity. The control circuit, through sensorless inference, calculates in real time whether the heating voltage of the filament has reached the rated value. At the same time, this voltage data, along with the anode current, is sent to the recursive timing model with physical conservation constraints to begin deducing the reference temperature of the filament.

[0227] Steady-state switching decision (physical state deduction layer) (Core Control Decision Layer): After 15 seconds, the reference temperature reaches 1600 degrees Celsius. At this time, the frequency domain spatial mapping module calculates the temperature of the "coldest point" at both ends of the filament to be 1560 degrees Celsius. The system prepares to switch to energy-saving mode. The integral barrier safety switching controller substitutes the inherent 800 nanosecond total delay time of the hardware into the simulation model and predicts that the temperature of the "coldest point" will drop to 1505 degrees Celsius after the switch.

[0228] Security Execution and Monitoring (Core Control Decision Layer) (Underlying Hardware and Sensing Layer, Step 4): Because the predicted temperature exceeds the vibration cessation threshold of 1500 degrees Celsius, the quadratic programming solver determines it is safe. The control circuit immediately sends a switching pulse through the micro-transformer on the printed circuit board. The high-voltage side switch operates, and the filament voltage decreases. At the same time, the topology residual diagnostic module continuously monitors and confirms that the measured value on the low-voltage side is consistent with the state of the deduced model, proving that the switching was successful and there was no physical jamming or short circuit.

[0229] Example 2: Application of precision power control in semiconductor plasma etching equipment.

[0230] 1. Application Scenarios and Technical Requirements: In the plasma etching process for semiconductor wafers, the power stability of the microwave source directly determines the uniformity and yield of the etching. During the process, the microwave power needs to be pulsed at the millisecond level (e.g., a square wave output with a 50% duty cycle). At the falling edge of each pulse, the filament temperature drops instantaneously.

[0231] 2. Systematic and collaborative implementation process for each step: Offline tuning (system configuration layer, anomaly protection layer): For high-precision pulse magnetrons used in etching equipment, an offline tuning method is employed to lock in the optimal dynamic response inductor combination and delay compensation parameters. Simultaneously, a period-aware asymptotic gain algorithm accurately calculates the envelope of high-frequency transient spikes caused by pulse modulation, solidifying a bypass capacitor capable of completely absorbing them.

[0232] Predictive intervention based on the falling edge of the pulse (underlying hardware and sensing layer) Physical state deduction layer (Core Control Decision Layer): At the instant the pulse power drops, the control circuit senses the power surge through sensorless inference. The temperature field simulation model immediately predicts that the filament's "coldest point" temperature will drop below the emission threshold in 2 milliseconds.

[0233] Active preheating intervention (core control decision layer): The solver of the integral barrier safety switching controller determines that oscillation will stop if no intervention is provided. Therefore, as the power decreases, the controller actively triggers physical intervention: instead of switching the filament inductance, it reverses the adjustment of the anode power loop, forcibly injecting a small sustaining power (e.g., 5% of the rated power) during the low-power plateau period of the pulse to compensate for the "micro-heating" of the filament.

[0234] Smooth recovery of pulse rising edge (underlying hardware and sensing layer) Physical state deduction layer (Core Control Decision Layer): Before the next pulse rising edge arrives, the temperature field simulation model confirms that the filament has sufficient thermal reserves. The safety switching controller then smoothly restores the filament power supply to normal. Throughout the process, it ensures that the filament temperature always fluctuates within the safe envelope under precise pulse modulation.

[0235] Example 3: Application of fault early warning and safe shutdown in medical microwave tumor hyperthermia equipment.

[0236] 1. Application Scenarios and Technical Requirements: In microwave tumor hyperthermia surgery, stable microwave source output and absolute safety are paramount. A sudden shutdown or high-voltage arcing during treatment poses a serious threat to the patient. This embodiment emphasizes the synergistic effect of fault diagnosis and safety protection.

[0237] 2. Systematic and collaborative implementation process for each step: Anomaly Detection During Operation (Underlying Hardware and Sensing Layer, Anomaly Protection Layer): When the device is stably outputting 400 watts of microwave power, the switching relay on the high-voltage side experiences contact sticking and jamming due to long-term thermal fatigue. At this time, the control circuit plans to issue a "reduce filament voltage" command in the next control cycle. However, after the command is issued, the sensorless inference algorithm detects that the high-voltage state inferred from the low-voltage side has not changed at all.

[0238] Topology residual feature vector generation (anomaly protection layer): The topology residual diagnosis module compares the "expected state" (voltage should decrease) with the "inferred state" (voltage remains unchanged), generating residual vectors that significantly deviate from zero. .

[0239] Precise fault location and safe shutdown (abnormal protection layer): The control circuit queries the built-in physical mapping table to determine that the residual feature vector precisely corresponds to the "high-voltage side switch physical jamming" fault. The system immediately executes the highest priority safety protection: Instantly cut off all high-voltage drive signals.

[0240] Turn off the main power circuit to stop microwave output.

[0241] The user interface displays a clear fault code, "Error Code: E-HVS-01 (High-Voltage Switch Failure)," guiding maintenance personnel to directly replace the high-voltage switch without the need for complex circuit troubleshooting.

[0242] Example 4: Application of anti-grid disturbance in the power supply of the pre-injector of a particle accelerator for scientific research.

[0243] 1. Application Scenarios and Technical Requirements: Microwave sources used in the pre-injection stage of particle accelerators are extremely sensitive to power grid quality. Instantaneous drops or surges in grid voltage are directly transmitted to the high-voltage side, causing drastic changes in filament heating power and resulting in beam instability.

[0244] 2. Systematic and collaborative implementation process for each step: Power grid disturbance capture (underlying hardware and sensing layer): When the voltage of the laboratory power grid drops by 15% instantaneously due to the start-up and shutdown of large equipment, the input variables of the sensorless inference algorithm are... The rapid change led to the inferred equivalent filament voltage on the high-voltage side. It also decreased significantly at the same time.

[0245] Rapid temperature field extrapolation (physical state extrapolation layer): After receiving the input of a sudden voltage drop, the temperature field extrapolation model immediately calculates that the temperature of the filament's "coldest point" will rapidly approach the oscillation stop threshold.

[0246] Proactive Robust Compensation (Core Control Decision Layer): Robust Error Compensation Term for Integral Barrier Safety Switching Controller This plays a crucial role under these operating conditions. Due to the significant forecasting disturbances introduced by grid fluctuations, the compensation term... The value increased sharply, significantly tightening the safety constraint boundaries.

[0247] Forced power compensation execution (core control decision layer): The quadratic programming solver determines that the system is in a high-risk state. The control circuit immediately performs physical intervention: it forces the anode power loop to instantaneously increase the output power, using the enhanced electron backflush effect to offset the insufficient filament heating caused by the grid voltage drop, thereby maintaining the filament temperature above the safety line until the grid is restored, ensuring stable injection of the particle beam.

[0248] Example 5: Application of rapid adaptation in the power supply of phased array radar transmitter.

[0249] 1. Application Scenarios and Technical Requirements: A certain type of phased array radar needs to replace the magnetron in its transmitter power supply from model A (rated filament voltage 6.3 volts) to model B (rated filament voltage 5.8 volts), which has superior performance but completely different electrical characteristics. Traditional replacement would require several weeks of manual parameter matching and destructive testing.

[0250] 2. Systematic and collaborative implementation process for each step: Parameter Proposal and Compliance Review (System Configuration Layer): R&D personnel input the datasheet parameters of the B-type magnetron into the offline tuning host computer. The proposal module generates a search space containing hundreds of candidate inductor combinations. The physical compliance review module completes pure mathematical verification within seconds, automatically eliminating 70% of the dangerous parameter combinations that could lead to transient arcing or thermal runaway.

[0251] Multi-fidelity data fusion optimization (system configuration layer): Within the remaining safety parameter space, the host computer automatically ran thousands of circuit simulations (low-fidelity data); subsequently, it guided test engineers to simply replace the three most promising inductor combinations on the physical prototype for actual power-on testing (high-fidelity data).

[0252] Optimal Parameter Output and Virtual Verification (System Configuration Layer): After fusing the two types of data, the autoregressive model converged within minutes and output the optimal inductor physical ratio and control delay parameters for the B-type magnetron. Subsequently, these parameters were fed into a software-in-the-loop (SIL) testbed, simulating all extreme temperature and vibration conditions that the radar might encounter in a virtual environment, confirming the robustness of the control logic.

[0253] Final solidification: After less than half a day of fully automated process, the new firmware parameters for the B-type magnetron are generated and can be directly used for batch burning on the production line, realizing agile adaptation for hardware replacement.

[0254] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0255] If the integrated units in the other embodiments described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0256] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for adjusting the filament voltage of a high-voltage microwave power supply, characterized in that, Applied to a voltage regulation circuit, the voltage regulation circuit includes: The hollow miniature transformer uses double or multi-layer epoxy glass cloth copper-clad plates, and uses copper foil traces to draw planar spiral coils on the top and bottom layers to form the primary winding and secondary winding of the hollow miniature transformer. A control circuit is located on the low-voltage side of the hollow micro transformer; a transmitting modulation circuit is provided on the low-voltage side, and the transmitting modulation circuit is connected to the control circuit; a receiving demodulation circuit is provided on the high-voltage side of the hollow micro transformer. A switching circuit, wherein the control terminal of the switching circuit is connected to the demodulation circuit of the receiving end; An inductive component, comprising at least one inductor, is connected to the path terminal of the switching circuit and is configured to, when the switching circuit is operating, connect the power supply inductor on the high-voltage side of the microwave power transformer in series or in parallel using the at least one inductor to adjust the power filament voltage; the switching circuit and the inductive component are located on the high-voltage side; The voltage adjustment method includes: The control circuit acquires the instantaneous amplitude of the AC input voltage on the low-voltage side, the amplitude of the AC output current of the primary circuit of the microwave power transformer, and the instantaneous adjustment signal output by the current proportional-integral regulator inside the control circuit. The control circuit infers the equivalent filament voltage on the high-voltage side based on the instantaneous amplitude of the low-voltage side AC input voltage, the amplitude of the AC output current, and the instantaneous adjustment signal output; wherein, the control circuit is equipped with a recursive timing model; The control circuit inputs the equivalent filament voltage on the high-voltage side and the collected magnetron anode high-voltage DC current to the recursive timing model so that the recursive timing model outputs the predicted value of the average reference temperature of the filament. The control circuit determines the lowest temperature point along the filament axis based on the predicted average reference temperature of the filament. The control circuit determines whether to immediately release, delay execution, or modify the inductor control command based on the lowest temperature point and the constraint judgment constant; the inductor control command is used to control the operation of the switching circuit; the constraint judgment constant is used to characterize the physical limit temperature required for the magnetron to maintain stable microwave output. When the switching circuit is working, the control circuit uses the AC output current amplitude, the duty cycle command of the instantaneous adjustment signal, the equivalent filament voltage on the high-voltage side, and the modulation amplitude index to locate potential physical faults on the high-voltage side.

2. The method according to claim 1, characterized in that, The control circuit includes a synchronous quadrature demodulation unit. The control circuit infers the equivalent filament voltage on the high-voltage side based on the instantaneous amplitude of the low-voltage side AC input voltage, the amplitude of the AC output current, and the instantaneous adjustment signal output. This includes: The control circuit inputs the instantaneous adjustment signal to the synchronous quadrature demodulation unit, so that the synchronous quadrature demodulation unit multiplies the instantaneous adjustment signal with the grid synchronous quadrature signal generated by the phase-locked loop to obtain the first sequence and the second sequence. The control circuit inputs the first sequence and the second sequence into a first-order recursive digital low-pass filter to filter out the second harmonic component and extract the DC component, thereby obtaining the first DC component corresponding to the first sequence and the second DC component corresponding to the second sequence. The control circuit obtains the modulation amplitude index based on the first DC component and the second DC component; The control circuit outputs the inferred equivalent filament voltage on the high-voltage side based on the modulation amplitude index, the instantaneous amplitude of the low-voltage side AC input voltage, the amplitude of the AC output current, the inherent turns ratio of the secondary and primary sides of the microwave power transformer, and the leakage inductance and equivalent line resistance of the corresponding primary side of the microwave power transformer.

3. The method according to claim 2, characterized in that, The control circuit calculates the modulation amplitude index using the following formula: ； Among them, the This represents the first DC component. This represents the second DC component. This represents the modulation amplitude index.

4. The method according to claim 1, characterized in that, When the recursive time series model is calibrated offline, the weight matrix corresponding to the recursive time series model satisfies the following linear matrix inequality: ;in, Denotes the first auxiliary matrix. Denotes the second auxiliary matrix. Describes a positive definite symmetric Lyapunov matrix. An augmented transition matrix that includes the system matrix and the boundaries of nonlinear terms.

5. The method according to claim 1, characterized in that, The control circuit determines the lowest temperature point along the filament axis based on the predicted average reference temperature value of the filament, including: The control circuit constructs a three-dimensional input tensor based on the predicted average reference temperature of the filament over the past K time steps, the filament axial spatial coordinates, and the inductor control command to be executed. The control circuit performs frequency domain mapping and global integration on the three-dimensional input tensor to obtain the expected absolute temperature distribution curves of the filament from the bottom to the top. The control circuit performs minimum value optimization on the expected absolute temperature distribution curve and takes the minimum value as the lowest temperature point along the filament axis.

6. The method according to claim 5, characterized in that, The control circuit performs frequency domain mapping and global integration on the three-dimensional input tensor to obtain the expected absolute temperature distribution curves of the filament from bottom to top, including: The control circuit uses a local linear transformation to project the three-dimensional input tensor to a high-dimensional hidden feature space. The control circuit performs a discrete fast Fourier transform on the spatial domain features in the high-dimensional hidden feature space to obtain the frequency domain features. The control circuit performs low-pass truncation on the frequency domain features and multiplies them with a pre-calibrated complex frequency domain weight matrix to obtain the inverse transform back to the spatial domain features. The control circuit performs dimensionality reduction on the inverse transformation back to the spatial domain features to obtain the expected absolute temperature distribution curves of the filament from the bottom to the top.

7. The method according to claim 1, characterized in that, The control circuit determines whether to immediately release, delay execution, or modify the inductor control command based on the lowest temperature point and the constraint judgment constant, including: The control circuit solves for the minimum intervention amount based on the lowest temperature point and the constraint judgment constant. The control circuit responds to the presence of the minimum intervention amount, and the minimum intervention amount is less than the first threshold, by issuing the inductor control command; The control circuit triggers a physical protection intervention strategy in response to the absence of a minimum intervention amount or the minimum intervention amount being greater than a second threshold; wherein the first threshold is less than the second threshold. The physical protection intervention strategy includes: The control circuit stops issuing the inductor control command; Alternatively, before issuing the inductor control command, the control circuit controls the anode high-voltage power supply loop to increase the anode set power, and then issues the inductor control command. Alternatively, the control circuit waits for the minimum intervention amount to be calculated in the next control cycle until the minimum intervention amount is less than the first threshold, and then issues the inductor control command.

8. The method according to claim 1, characterized in that, When the switching circuit is operating, the control circuit uses the AC output current amplitude, the duty cycle command of the instantaneous adjustment signal, the equivalent filament voltage on the high-voltage side, and the modulation amplitude index to locate potential physical faults on the high-voltage side, including: When the switching circuit is working, the control circuit calculates the first residual based on the amplitude of the AC output current, the duty cycle command of the instantaneous adjustment signal, and the equivalent filament voltage on the high-voltage side. When the switching circuit is working, the control circuit calculates the second residual based on the AC output current amplitude, the modulation amplitude index, and the high-voltage side equivalent filament voltage. The control circuit determines that the switch contacts in the switch circuit are stuck due to sintering when the absolute value of the first residual is greater than the third threshold and the absolute value of the second residual is within a preset range of 0. The control circuit determines that an inductor in the inductive component has experienced an inter-turn insulation breakdown short circuit when the absolute value of the first residual is greater than the third threshold and the absolute value of the second residual is greater than the third threshold.

9. The method according to claim 1, characterized in that, The method further includes: When the system is configured at the factory or dynamically adjusted, the control circuit uses the periodic disturbance time constant of the rated grid input of the microwave power supply, the upper limit amplitude of the DC bias voltage, the limit of the high-frequency AC disturbance amplitude, and the equivalent parasitic inductance and impedance parameters determined by the layout of the double-layer or multi-layer epoxy glass cloth copper-clad board to calculate the optimal capacitance value of the high-voltage side auxiliary power supply bypass absorption capacitor of the hollow micro transformer.

10. The method according to claim 1, characterized in that, The method further includes: Build a parameter proposal module and a physical compliance review module in the host computer; The host computer uses the parameter proposal module to generate a search subspace containing several candidate inductor combinations based on historical security configurations and boundary feedback summaries from the previous test. The host computer uses the physical compliance review module to trim several candidate inductor combinations in the search subspace to obtain a safe search subspace containing several target inductor combinations that meets the requirements. The host computer calls the underlying circuit to perform automated simulation of the target inductor combinations in the safe search subspace, and obtains the simulation score corresponding to each target inductor combination. And select a preset number of target inductor combinations from the simulation scoring to conduct power-on tests to obtain the actual score; The required inductor combination and the optimal delay compensation time are determined based on the simulation score and the actual score.

Images