Surveillance en ligne du vieillissement des condensateurs dans les alimentations laser

Surveillance en ligne du vieillissement des condensateurs dans les alimentations laser

Abstract

Laser marking supplies fail either suddenly or gradually, and only the gradual failures give warning. This article gives the four measurable aging signatures, the online ESR extraction from ripple data, the PD precursor that opens the widest intervention window, four reasons monitoring programs fail, six implementation strategies, and the statistical alert rules that keep a monitor trusted.

1. Introduction

Laser marking machines fail in two ways. Some parts short out with no warning at all, and the line stops until a replacement arrives. Others drift for months, follow a measurable trend, and can be caught weeks before they cross a limit.

Partial discharge, capacitance drift, ESR growth and leakage rise all leave traces in the machine's own operating data. A sensor and an average over a few hundred pulse periods turn those traces into a trend.

The power supply is the richest source of those traces, because its capacitors age faster than the control electronics. The failure that stops the line is almost always a capacitor, and the trend that predicts it is already present at the switch node.

This article gives the aging signatures, the online measurement arithmetic, four reasons monitoring programs fail, six strategies that make them work, and the alert rules that keep the data readable.

2. Mécanisme de base

2.1 The four aging signatures

Capacitor degradation shows up in four measurable parameters, and each trend is its own diagnostic.

C(t), ESR(t), I_leakage(t), PD_activity(t)

where C(t) is the capacitance trend, which drifts downward as the dielectric ages, ESR(t) is the series resistance trend, which grows as the electrode interfaces degrade, I_leakage(t) is the leakage-current trend, which rises with contamination and moisture, and PD_activity(t) is the partial-discharge magnitude and count trend.

The classic failure-prediction shape is a slow drift followed by an accelerating knee.

P(t) = P_0 × [1 + α t + β exp(γ (t − t_crit))]

where P(t) is the parameter value, P_0 is its baseline, α is the slow linear rate, β and γ set the knee amplitude and steepness, and t_crit is the time at which the knee begins. Before t_crit the monitor watches a trend; after it the monitor schedules an intervention.

2.2 Extracting ESR from ripple data

The most useful online metric is ESR, and it comes from the ratio of ripple voltage to ripple current at the capacitor.

ESR ≈ ΔV_ripple / ΔI_ripple

where ΔV_ripple is the sampled ripple voltage across the capacitor and ΔI_ripple is the ripple current measured by a shunt or a current transformer. The controller computes the ratio in real time.

A 4.7 nF part at 10 kV that carries 2 A rms through 125 mΩ of ESR shows 250 mV of ripple. The same part at 250 mΩ shows 500 mV at the same current, and that jump is the whole signal the monitor needs.

P_loss,new = P_loss,old × (ESR_new / ESR_old)

where P_loss,new and P_loss,old are the self-heating before and after the change, and ESR_new and ESR_old are the two resistance readings. Self-heating tracks the resistance one to one, so a 2× ESR rise doubles the heat and marks end of life. A 30–50 % rise is the early-warning band, not the flag.

2.3 Partial discharge as the high-voltage precursor

On the high-voltage stages, partial discharge is the earliest warning, and incipient activity precedes visible flashover by months. A coupling capacitor and a high-frequency current transformer on the Q-switch output measure the apparent charge.

q_apparent = ∫ I_pd(t) dt

where q_apparent is the apparent charge in picocoulombs, I_pd(t) is the discharge current waveform, and the integral runs over the discharge event. A rising trend in apparent charge or in discharge count at constant operating voltage points to void growth, and the intervention window opens weeks before the failure.

A part that holds 5 pC at 10 kV and starts reporting 15 pC at the same voltage and the same 20 kHz duty has begun to degrade. The 20–40 kV stages are where this reading pays for itself.

2.4 Baselines and alert rules

Every monitor needs a baseline: parameter values recorded at commissioning and normalized to temperature, load and repetition rate.

Alert if P(t) > P_base × 1.5 OR dP/dt > 3σ

where P(t) is the current reading, P_base is the baseline median, and σ is the standard deviation of the historical noise on that channel. The thresholds must come from the machine's own history, because every machine runs a different profile.

Normalization matters more than the rule. A Class I part at ±30 ppm/°C contributes 0.06 % of drift over a 20 °C swing, so the residual trend is genuine aging. A Class II part contributes 2–5 % over the same swing and produces false alerts every season.

Fig. 1 — A Class II filter reaches the 150 % ESR alert near month 44, while a Class I part stays inside 118 % after five years of service.

3. Points faibles en matière d'ingénierie

Pain Point 1: Baselines that were never recorded

The machine was commissioned without a healthy-state record. Two years later the monitor reports a high ESR and nobody can say whether it is high or normal. Without a baseline the trend has no meaning, and the monitor is switched off within a month.

Pain Point 2: Sensor noise swamping the signal

The ripple-voltage channel picks up switching noise, and the computed ESR bounces ±50 % between samples. The trend sits inside the noise, the thresholds become arbitrary, and the engineers stop trusting the numbers. Filtering and synchronous averaging are prerequisites.

Pain Point 3: Monitoring the parts without the conditions

The monitor logs ESR but not cabinet temperature, so the seasonal swing looks like drift. False trends trigger needless replacements, the team loses faith in the system, and the real trend stays hidden until temperature normalization is added.

Pain Point 4: Alerts without a workflow

The system detects a rising discharge trend and no action follows: no spare-part trigger, no scheduled window, no escalation path. The alert is logged and ignored, and the capacitor fails two weeks later. The information existed; the process did not use it.

4. Étapes de conception

Strategy 1: Capture commissioning baselines

Record capacitance, ESR, leakage and PD for each capacitor group at commissioning, normalized to operating conditions, and store the values in the machine health record. That baseline is the most valuable data the monitor will ever hold.

Strategy 2: Normalize every reading to temperature and load

Log cabinet temperature and load alongside each reading, and normalize the trends with the component's own TCC and TCR figures. Specify N4700 Class I parts at ±30 ppm/°C, where the residual trend is aging and almost nothing else.

Strategy 3: Beat the switching noise with synchronous averaging

Sample the ripple voltage in step with the switching cycle, average over hundreds of pulse periods, and low-pass filter the ESR estimate. A stable trace with a noise band below 5 % of the mean is the precondition for thresholds anyone will trust.

Strategy 4: Fit a PD port at design time

Add a coupling capacitor and a high-frequency current transformer to the Q-switch output while the board is being laid out. A quarterly PD reading with a documented trend costs almost nothing at that point and covers the 20–40 kV stages.

Strategy 5: Give every alert level an action

Define three levels: trend rising means inspect at the next service window, threshold crossed means order the spare and plan the swap, and critical means intervene within 24 hours. Feed the alerts into the site maintenance system so the reading becomes a work order. A condensateur à film bank in a DC link follows the same rule.

Strategy 6: Specify parts with documented aging data

Choose capacitors with bounded capacitance drift, stable ESR, PD below 5 pC and published life data. Predictable parts make monitoring a measurement; unpredictable parts make it a lottery.

Paramètre de composant	Valeur typique	Condition de test
Tension nominale (série HVCT8G)	10 kV DC	haute tension 1.5× pendant 60 s
Capacitance	4700 pF ±20% (472M)	1 MHz, 1 V efficace
ESR à 1 MHz	inférieur à 50 mΩ	25 ° C
Inductance de boucle (accord E40)	en dessous de 8 nH	Espacement des bornes de 40 mm
La resistance d'isolement	supérieure à 10 GΩ à 500 V CC	25 °C, 60 % HR
Plage de température	−40 °C à +125 °C	Série IEC 60068-2
Capacité dv/dt d'impulsion	au-dessus de 50 kV/µs	à la tension nominale

5. HVC capacitors for monitorable aging

Three properties make aging readable: a small temperature coefficient, a stable loss factor, and a documented PD floor. HVC builds N4700 parts for all three.

Six paramètres importants ici

• ±30 ppm/°C contributes 0.06 % of capacitance drift over a 20 °C swing, so the monitored trend comes from aging and not from the cabinet.
• tanδ at or below 0.1 % keeps ESR flat as the body heats, so a rising ESR reading means degradation.
• PD below 5 pC at rated voltage, with full-lot verification, gives the PD monitor a clean floor: a rise measured in the field is real.
• 1 kV to 100 kV coverage lets one monitoring method span the whole supply, from the Q-switch driver to the 40 kV stage.
• Documented life and aging data support baseline and threshold engineering, and the data ships with the parts.
• Field record: an HVC 40 kV / 441 HVCT8G-class screw-type capacitor qualified as a drop-in replacement for Japanese Murata parts in lithography light-source field repair, and the same construction matches the electrical parameters of références croisées candidats.

Livraison et support

Les composants standard sont expédiés sous 4 à 8 semaines et les modules appariés sous 6 à 10 semaines. Au-delà de 1 kV, le site est disponible en module série avec appariement. résistances de gradation.

Send the monitored parameters, the sensor positions and the operating profile. HVC returns expected aging envelopes, ESR-versus-temperature data and a threshold table, plus a clamping diode à récupération rapide for the high-voltage node.

6. Questions fréquentes

Q1: Which parameter should the monitor trend first?

ESR. It rises before capacitance moves, it comes from ripple data the controller already samples, and self-heating tracks it one to one, so a 2× rise doubles the heat inside the part.

Q2: At what ESR rise is the capacitor at end of life?

A 2× rise over baseline, which doubles the ripple self-heating. A 30–50 % rise belongs to the early-warning band and belongs in the trend, not in the replacement decision.

Q3: How early can partial discharge warn of a failure?

Months. Incipient discharge activity grows long before visible flashover, and a quarterly reading on the 20–40 kV stage with a documented trend is the widest intervention window available.

Q4: Why normalize the reading to temperature?

Because a Class II part moves 2–5 % over a 20 °C seasonal swing, which reads as drift. A Class I part at ±30 ppm/°C moves 0.06 % over the same swing, so the residual trend is aging.

Q5: How stable does the ESR reading need to be?

A noise band below 5 % of the mean. Synchronous sampling, averaging over hundreds of pulse periods and a low-pass filter get there; without them the trend sits inside the noise.

Q6: What turns a monitoring system into a maintenance system?

Three defined actions, one per alert level: inspect at the next window, order the spare, or intervene inside 24 hours. The reading has to arrive as a work order, not as a log line.

7. Références

1. E. Kuffel, WS Zaengl, J. Kuffel, Principes fondamentaux de l'ingénierie haute tension, 2e éd., Newnes, 2000.
2. N. Mohan, TM Undeland, WP Robbins, Électronique de puissance : convertisseurs, applications et conception, 3e éd., Wiley, 2003.
3. Avis MIL-HDBK-217F 2, Prédiction de la fiabilité des équipements électroniques 1995.
4. CEI 60384-1:2021, Condensateurs fixes pour équipements électroniques - Partie 1 : Spécifications générales, 6e éd. Genève : CEI, 2021.
5. CEI 60270:2025, Techniques de test haute tension - Mesure des décharges partielles basée sur la charge, 4e éd. Genève : CEI, 2025.
6. GB/T 7354-2018, 高电压试验技术局部放电测量. Publié : 中国标准出版社, 2018.
7. B. Wang, J. Meng et P. Zhao, Surveillance du vieillissement des condensateurs électrolytiques en aluminium dans les variateurs de vitesse, IEEE Transactions on Power Electronics, vol. 37, no. 4, pp. 4564-4574, 2022. https://doi.org/10.1109/tpel.2021.3121813.
8. H. Nakao, Y. Yonezawa, T. Sugawara, Y. Nakashima et F. Kurokawa, Méthode d'évaluation en ligne de la dégradation des condensateurs électrolytiques pour la prédiction des défaillances des alimentations à découpage à commande numérique, IEEE Transactions on Power Electronics, vol. 33, no. 3, pp. 2552-2558, 2018. https://doi.org/10.1109/tpel.2017.2691048.
9. Z. Li et al., Méthode de surveillance intelligente de l'état de dégradation des condensateurs à film métallisé par le signal ultrasonique de la décharge auto-cicatrisante, dans les actes du Symposium international CPSS & IEEE sur le stockage et la conversion de l'énergie (ISESC) 2024, p. 206-211, 2024. https://doi.org/10.1109/isesc63657.2024.10785409
10. États-Unis 6,937,027, Capteur hybride pour la détection des décharges partielles à haute fréquence, Hanyang Hak Won Co., Ltd., délivré en 2005, https://www.freepatentsonline.com/6937027.html.
11. États-Unis 10,529,487, Condensateur céramique multicouche avec électrodes internes présentant différentes positions de bord, Taiyo Yuden Co., Ltd., accordé en 2020, https://www.freepatentsonline.com/10529487.html .
12. Équipe d'ingénierie des applications HVC, Condensateur céramique haute tension Ressources techniques, https://www.hv-caps.com