可变频率等离子体点火负载自适应技术可减少失火

可变频率等离子体点火负载自适应技术可减少失火

抽象

A gas gap, a torch cable and a gas recipe that all change between jobs turn a fixed ignition frequency into a compromise. This article derives the tank resonance, the phase-zero condition, the loop Q, the blanking window and the energy governor for a 4700 pF ignition stage, then gives four pain points, six design steps and the capacitor data that keeps the frequency-to-load map of the algorithm valid.

1. 引言

A plasma torch is a moving target for the ignition circuit. The gas gap changes as the electrode wears, the gas mix and flow follow the cutting recipe, cable length changes when the machine is reconfigured, and the whole ignition chain warms up across a shift. A fixed-frequency, fixed-energy igniter treats every one of those states as identical.

The cost lands in mixed-material shops. A machine tuned on clean dry air and mild steel misbehaves on aluminium with nitrogen, or on rusty plate in a humid shop. Each misfire spends time, nozzle life and operator attention, and no measurement records why it happened.

Robotic cells feel it harder. One misfire starts a retry sequence, and repeated retries abort the whole nesting program. Field data shows fixed-tuned igniters misfiring on 3 to 8% of attempts under varied conditions and adaptive systems holding below 1%. Over a 120-pierce nesting program, a 3% rate leaves 97 programs in 100 with at least one misfire; 0.9% still leaves 66.

This article sets out the adaptation loop in numbers: the tank resonance as the load moves it, the phase-zero condition that the controller tracks, the loop Q that sets the sweep step, and the energy governor that keeps breakdown reliable without spending consumable life.

2. 核心机制

2.1 Tank resonance as the load moves it

The tank capacitor and the ignition loop inductance set the frequency at which the drive couples best into the gap.

f_r = 1 / (2 π √(L × C))

where f_r is the resonant frequency, L is the total loop inductance and C is the tank capacitance. The unit label needs care: with L in microhenries and C in picofarads the working form is f_r = 159.15 / √(L × C) in MHz, and with L in nanohenries the constant becomes 5033. Entering the raw numbers with no constant returns a value 31,600 times too small for megahertz.

With 1.5 µH and 4700 pF the tank rings at 1.896 MHz in a bare loop. A torch cable adds stray capacitance in parallel: 3 m of cable and a head assembly bring 470 pF, which lowers the effective resonance to 1.807 MHz, and a 6 m cable brings 770 pF and 1.757 MHz. The load moves the useful band by 140 kHz before the gap state is even considered.

2.2 Damping and the phase-zero condition

Resistance in the loop lowers the resonance, and the drive sees zero phase between tank voltage and tank current at that point.

ω_res = √(1 / (L × C) − (R_total / (2L))²)

where ω_res is the damped angular resonant frequency, R_total is the total loop resistance, L is the loop inductance and C is the tank capacitance.

At 200 mΩ of loop resistance the damping term reaches 4.4 × 10⁹ against 1.4 × 10¹⁴ for the LC term, so it changes the resonance by less than 0.02%. Damping shifts the frequency very little and the shape a great deal. That is why the controller tracks phase, and leaves amplitude to the energy governor.

2.3 Loop Q and the sweep step

The sharpness of the resonance decides how far the drive may move in one correction step.

Q = (1 / R_total) × √(L / C)

where Q is the quality factor of the ignition loop, R_total is the loop resistance, L is the loop inductance and C is the tank capacitance.

The same 1.5 µH and 4700 pF give a characteristic impedance of 17.9 Ω. A cold gap with 0.6 Ω in the loop returns Q = 30 and a 60 kHz band; a pre-ionized gap raising the loop to 1.5 Ω returns Q = 12 and a 146 kHz band. The phase slope follows the band, so the same correction step lands very differently on the two states.

2.4 The phase-locked correction law

The controller measures phase at each cycle and corrects the drive frequency with proportional and integral terms.

Δω_next = k_p × φ_measured + k_i × ∫ φ_measured dt

where Δω_next is the next frequency correction, φ_measured is the measured phase error between drive voltage and tank current, and k_p and k_i are the proportional and integral gains.

A gain of 114 kHz per radian converts a 10° error into a 20 kHz step. At 1.9 MHz that is a 1.05% move, which covers the 140 kHz load band in seven steps and stays inside the 60 kHz band of a cold gap. Blanking sets the fastest usable loop: the drive period at 1.9 MHz reaches 527 ns, so a blanking window of 1 to 2 µs would swallow two to four whole cycles and freeze the estimate. Hold the window at or below 50 ns.

2.5 The energy governor

The second half of the adaptation raises energy until the gap breaks down, inside a cap set by the recipe.

E_ign = E_nominal + g(material) × ΔE_latency

where E_ign is the energy delivered on the attempt, E_nominal is the energy for a healthy gap, ΔE_latency is the energy added per latency step and g(material) is the recipe gain reflecting how much extra energy the workpiece tolerates.

Each 1 kV step on a 4700 pF tank adds 49 mJ, which is 21% on the 0.235 J base. A 30 µs latency target with 1 kV steps reaches 0.333 J after two steps, and the governor stops there for thin sheet.

2.6 Loop loss per strike

What the loop resistance costs in heat follows the square of the current.

P = I² × R

where P is the instantaneous dissipation, I is the loop current and R is the loop resistance. Amperes squared with milliohms return milliwatts, which is the second unit trap in this stage: 480 A peak into a 600 mΩ cold loop reads as 138,240,000 and means 138 kW at the current peak.

The root-mean-square current over the 264 ns half-cycle reaches 339 A, so that loop dissipates 18.2 mJ per strike, which is 7.7% of the 0.235 J stored. Of that, the 50 mΩ equivalent series resistance of the capacitor accounts for 1.5 mJ and the rest leaves through the transformer and the wiring.

Fig. 1 — The tank phase crosses zero at 1.807 MHz with a cold gap and 1.757 MHz pre-ionized; the loop sweeps the 140 kHz band between them.

3. 工程痛点

Pain Point 1: Sweep speed against resonance sharpness

The loop has to find the new resonance before the gap moves again, and a sharp resonance is narrow. A cold gap with Q 30 spans 60 kHz, so a 20 kHz correction step works; the same step overshoots once pre-ionization broadens the response to 146 kHz. Gains have to follow the measured Q, and Q itself moves with gap state.

Pain Point 2: Phase measurement inside ignition noise

Tank voltage and current carry ignition EMI and power-stage switching transients, so the comparator needs filtering and blanking. The filters bring their own danger: a frequency-dependent delay of 20 ns at 1.9 MHz is 13.7° of phase, which the loop reads as a real frequency error and corrects toward the wrong point.

Pain Point 3: Capacitor drift corrupts the frequency-to-load map

When the tank capacitance drifts with temperature or bias, the controller corrects for the capacitor and reports a load change that never happened. A 4700 pF part moving 1% is 47 pF, which sits in the same range as the 470 pF of cable and head capacitance the loop exists to track.

Pain Point 4: Energy governor tuned without consumable data

The cap on ignition energy balances breakdown reliability against ablation rate. Without strikes-per-set telemetry the governor is set by opinion: too low and the machine misfires under humid conditions, too high and the electrode tip pays for every pierce on thin sheet.

4. 设计步骤

Strategy 1: Sweep on phase with gains scheduled from Q

Measure tank phase continuously with a fast comparator, blank 50 ns or less around each drive edge, and schedule k_p and k_i from the measured Q so the loop stays critically damped in both gap states. A 114 kHz per radian gain gives a 20 kHz step on a 10° error, which covers 140 kHz in seven steps.

Strategy 2: Bound the sweep band from the hardware

Choose tank components so the whole useful band, load shifts included, sits inside the oscillator range with 20% margin. On a 1.896 MHz bare tank that means a range of 1.55 to 2.25 MHz to cover the 1.757 to 1.896 MHz load band and still leave headroom for a hot tank.

Strategy 3: Make the governor recipe-driven

Derive the energy cap from the material and gas entries of the cutting program, and bias E_ign inside that cap from the measured breakdown latency. Update the cap monthly from strikes per set so the machine learns where thin sheet starts costing consumable life.

Strategy 4: Log the adaptation state on every attempt

Record frequency, phase error, energy used and breakdown latency. A persistent frequency shift points at a load change such as electrode wear, a re-routed cable or a gas change, and it should raise a maintenance check on the torch, not a firmware retune.

Strategy 5: Degrade through a retry sequence

When the first attempt fails, change one variable on each retry: search at the band edge, then raise energy, then extend duration. Repeating the same failed attempt spends 0.235 J per try and teaches the loop nothing about which part of the load space it is missing.

Strategy 6: Validate on a programmable load simulator

Build a dummy torch from a switched resistor and capacitor ladder that emulates gap states from cold to pre-ionized, then verify that the loop converges inside one ignition window across the full matrix before any field trial. Record the frequency each state settles at and compare it with the model.

组件参数	典型值	测试条件
额定电压（HVCT8G 系列）	10 kV DC	高压1.5倍，持续60秒
电容量	4700 pF ±20% (472M)	1 MHz，1 V rms
1 MHz 下的 ESR	低于 50 mΩ	-
自感	低于 8 nH	端子间距 40 毫米
温度系数	±30 ppm/°C	−40 °C 至 +125 °C
绝缘电阻	500V直流电压下电阻大于10GΩ	25℃，60%相对湿度
局部放电	低于 5 摄氏度	额定电压

5. HVC N4700 (T3M) capacitors for adaptive ignition

The algorithm stands on a tank whose frequency-to-load map holds. HVC builds the N4700 (T3M) Class I family for that role, with the 厚膜电阻 和 采样电阻 needed to sense the tank current, from a 6,000 m² owned factory in Dongguan across a 1 kV to 100 kV range.

这里需要考虑的六个参数

• Stable tank capacitance. A ±30 ppm/°C temperature coefficient keeps the frequency-to-load map valid from a cold start to a hot cabinet, so the loop tracks the torch and not its own capacitor.
• Low loss for a measurable resonance. A dissipation factor below 0.1% keeps the tank Q high and the phase signal clean, which is what the comparator needs in an ignition environment full of switching transients.
• Behaviour that holds over life. Partial discharge below 5 pC and the Class I formulation keep the algorithm tuned at commissioning valid after millions of pulses, with no aging drift to absorb.
• Low parasitic inductance in compact packages. A self-inductance below 8 nH and a short mounting loop keep the ring at 1.9 MHz, so the usable sweep band stays wide and the phase measurement keeps its resolution.
• Coverage of the whole platform. Ratings from 1 kV to 100 kV let one adaptive design serve portable igniters and heavy-plate injectors with the same firmware and the same tuning method.
• Proven stability in adaptive systems. The 40 kV 441-series doorknob capacitors replaced first-tier brand parts in high-voltage field-service programs, and a 交叉参照 against those parts shows the drift a control loop would have to absorb.

交付和支持

Standard parts ship in 4 to 8 weeks and matched tank sets in 6 to 10 weeks. HVC applications engineering supports resonance survey, gain scheduling, blanking-window sizing and dummy-load test design. Send the ignition frequency range, tank voltage and load-variation profile, and HVC returns a tank recommendation with PD data and sweep-band calculations within five working days.

6. 常见问题

Q1: Why does a fixed ignition frequency miss?

Tank resonance follows the load. A 4700 pF tank with 1.5 µH rings at 1.896 MHz bare, 1.807 MHz with 3 m of cable and 1.757 MHz with 6 m. A drive parked at one frequency sits off resonance on most of those states, so the coupling drops and breakdown takes longer or fails.

Q2: How does the controller find the resonance?

It measures the phase between drive voltage and tank current. Phase crosses zero at resonance, so the correction can be proportional and integral on the phase error alone. A 114 kHz per radian gain turns a 10° error into a 20 kHz step and covers the 140 kHz load band in seven steps.

Q3: How wide should the blanking window be?

At 1.9 MHz the drive period is 527 ns, so a 1 to 2 µs blanking window removes two to four whole cycles and stalls the estimate. Set the window at 50 ns or less, which is about 10% of the period, and compensate the comparator delay in firmware.

Q4: Why does loop Q matter to the algorithm?

Q sets the phase slope, and the slope sets how far one correction step may move. A cold gap at Q 30 has a 60 kHz band, so a 20 kHz step is safe. A pre-ionized gap at Q 12 has a 146 kHz band, and the same step is conservative. Gains scheduled from Q keep the loop critically damped in both states.

Q5: How much energy does a retry add?

Each 1 kV step on a 4700 pF tank adds 49 mJ, or 21% on the 0.235 J base. Two steps take the attempt to 0.333 J. Cap the sequence at the recipe value for the material, because the extra energy lands on the electrode tip while the arc root is still narrow.

Q6: Which capacitor suits an adaptive ignition tank?

An N4700 Class I part with a dissipation factor below 0.1%, a ±30 ppm/°C coefficient, self-inductance below 8 nH and partial discharge below 5 pC at rated voltage. Those numbers keep the resonance where the algorithm expects it across temperature, bias and pulse count.

7。 参考

1. E. Kuffel、WS Zaengl、J. Kuffel 高电压工程基础知识，第 2 版，Newnes，2000 年。
2. Mohan, N.、Undeland, TM 和 Robbins, WP， 电力电子学：变换器、应用和设计，第 3 版。霍博肯，新泽西州：约翰·威利父子出版社，2003 年。
3. MIL-HDBK-217F 通知 2， 电子设备可靠性预测1995。
4. IEC 60384-1:2021， 电子设备用固定电容器 - 第 1 部分：通用规范，第 6 版。日内瓦：IEC，2021 年。
5. IEC 61071:2017， 电力电子电容器，第 2.0 版。日内瓦：国际电工委员会，2017 年。
6. A. Boudiaf、S. Bouazabia、N. Harid 和 ML Amrani， 高压电缆绝缘层内气腔局部放电阈值的解析计算，电气工程，第 104 卷，第 2 期，第 555-565 页，2021 年。https://doi.org/10.1007/s00202-021-01310-9。
7. V. Belyaev， 真空绝缘体表面闪络模型，载于 ISDEIV 会议论文集。第 18 届真空放电和电绝缘国际研讨会，第 151-154 页，1998 年。https://doi.org/10.1109/deiv.1998.740596。
8. 美国 6,760,215, 具有高电压击穿阈值的电容器DF DeVoe（与 Presidio 相关），2004 年授权，https://www.freepatentsonline.com/6760215.html。
9. 驱动未来：高压MLCC如何提升现代电子产品的效率，Murata，https://www.murata.com/en-eu/s/blog/powering-the-future-how-high-voltage-mlccs-drive-efficiency-in-modern-electronics.html。
10. GB/T 7354-2018， 高电压试验技术局部放电测量。 北京: 中国标准出版社, 2018.
11. HVC应用工程团队 高压陶瓷电容器 技术资源https://www.hv-caps.com