Blogarticle
Test bench for lifetime testing: Realistic loads for greater reliability
The lifetime of power semiconductors determines the reliability of converters in wind turbines, industrial drives, or electromobility. An early failure can cause not only high costs but also long downtimes. To prevent this, semiconductors have so far been designed with conservative safety margins – which, however, leads to higher investment costs.
Thanks to an innovative power cycling test bench, we can offer our customers realistic lifetime tests. This allows for a more accurate assessment of the reliability of power semiconductors and more efficient component design.
What makes our test bench special?
Conventional lifetime tests mostly focus on conduction losses – they generate temperature cycles by operating components under constant high current. That is far from reality. In real applications, the stresses also result from switching losses and from the superposition of temperature fluctuations with different frequencies.
Our test bench integrates both:
-
Switching losses realistically reproduced – through targeted control of the power semiconductors and additional inductances in the current path.
-
Superimposed temperature ripples – fast and slow fluctuations occur simultaneously, as they do for example in wind power plants due to power variations and the fundamental frequency of the current.
Figure 1 illustrates a typical temperature pattern: fast fluctuations are superimposed with slower ones, creating a stress condition that comes much closer to actual operation.
Fig 1. Possible temperature profile with test bench
The profile consists of two superimposed temperature fluctuations. A classical rainflow algorithm will count the large fluctuation as the peak-to-peak value TPP. A calculation with low resolution for slow temperature fluctuations, however, would rather capture the mean-to-mean value TMM. This raises the question of which temperature fluctuation is actually relevant for lifetime and how the prevailing frequencies influence this. A particular advancement lies in the ability to generate multiple temperature ripples simultaneously. Instead of testing different load profiles one after the other, we can now combine them into a single pattern.
Fig 2. Pattern design: the simultaneous superposition of multiple temperature swings
A particular advancement lies in the ability to generate multiple temperature ripples simultaneously. Instead of testing different load profiles one after the other, we can now combine them into a single pattern.
Figure 2 illustrates this principle: two individual patterns – a fast temperature cycle of 50 K and a slower cycle of 72 K – are not applied separately over time but superimposed simultaneously. This creates a new, more realistic load profile.
This approach is crucial since, in real applications, many influences act at the same time. In this way, lifetime tests can be conducted closer to real operating conditions, reducing uncertainties in prediction.
How do these new load patterns affect the actual lifetime?
For this purpose, we use a Weibull analysis, which is well established in reliability engineering. Figure 3 shows the results of the initial lifetime experiments. Here, the spread of the EoL times for pattern A is clearly visible. This spread leads to a relatively low b-factor, i.e. the slope in the probability plot. As a result, the range of the 90% confidence intervals is also larger compared to the other results.
Fig 3. Weibull Plots for different Pattern A and for B & C
For pattern C, we see that all EoL results lie outside the confidence intervals of pattern B – and this even with the additional temperature ripple from pattern A. In other words, the average lifetime of the modules with pattern B (only one slow temperature ripple) was shorter than with the superposition (one slow temperature ripple plus fast temperature ripples). A classical rainflow analysis would simply sum up the lifetime consumption of both ripples; mathematically, pattern C should therefore reach end of life the fastest.
Fig 4. Lifetime of the different patterns, normalized to the average lifetime of pattern C
Figure 4 puts the results of all tests into perspective. The lifetimes are normalized to the mean lifetime of pattern C. The mean lifetime of pattern A was about 1.2 times that of pattern C. Pattern B reached about 0.8 times the lifetime of pattern C.
All specimens of pattern C had a longer lifetime than any specimen of pattern B – even despite an additional temperature ripple (50 K) on top of the fluctuation of pattern B (72 K).
Assuming a linear damage accumulation – that is, the way the rainflow algorithm would count the fluctuations in pattern C – a shorter lifetime would have been expected. A simplified estimation of this expected lifetime (LPC,expec.) of pattern C is given by the inverse of the sum of the lifetime consumption of patterns A and B:
Want to know more?
Contact: Get in touch
Publication: T.-M. Plötz, J. Fuhrmann and H.-G. Eckel, “Power Cycling Test Bench with Realistic Loss Distribution and Temperature Ripples,” 2022 24th European Conference on Power Electronics and Applications (EPE’22 ECCE Europe), Hanover, Germany, 2022, pp. 1-10.