InsightFab
Knowledge Base/System Reliability Simulation: Applying the Monte Carlo Method
Reliability6 min read

System Reliability Simulation: Applying the Monte Carlo Method

This article highlights a practical approach to understanding system reliability, drawing from the author's past experience with plummeting product yields and urgent inquiries about expected product lifespan. It emphasizes that product issues often arise from a lack of understanding of 'system reliability' rather than isolated component failures, akin to a computer system crashing due to poor component interaction despite sound individual parts. The content advocates for a system-level approach to reliability in product development, crucial for avoiding costly detours.

That Day, Yield Plummeted Terribly, My Boss Asked Me "What's the Expected Lifespan?"

I still remember two years ago, when our new process products were just mass-produced. Initially, the yield was acceptable, with a Cpk of around 1.08. However, not long after, customers started complaining that the products often failed halfway through use. What was worse, this failure mode wasn't something detectable early on; it would only manifest after accumulating a certain amount of time. My boss was furious, his face green, and he called me over, immediately asking: "Tell me, what is the 'expected lifespan' of this batch of products? Can we last until the next revision?" At that moment, I honestly had no idea; I just felt a chill down my spine.

Where Was the Problem? Not a Single Event, But a Chain Reaction!

Frankly speaking, the problem you're encountering is the same as what I faced that day: system reliability. Many people might think, "I've validated every component, and the Cpk targets were met, so why are there issues when they're assembled?" To be honest, it's like buying a bunch of parts to build a computer at home: the power supply is stable, the memory is stable, and the motherboard is stable, but when everything is put together, it might crash intermittently because of "interaction" issues between certain components, or environmental conditions that weren't considered.

Products on our production line usually have more than one failure mode. It could be high-temperature accelerated aging, moisture ingress, or vibration leading to solder joint fatigue. Each failure mode has its own "life distribution." When these failure modes coexist, and they are not completely independent, merely looking at the Cpk of individual components is simply not enough. You need a method to simulate the superposition of these "uncertainties" and understand what impact this will have on the overall product life. This is where the Monte Carlo method comes into play.

How Is It Actually Done? Using Random Simulation to Understand Life

So, the key point is that the Monte Carlo method is a "random simulation" approach. It doesn't calculate a precise value but simulates thousands of different scenarios and then observes the distribution of results from these scenarios.

For example, let's assume our product has two main failure modes:

  1. Mode A: Mean life of 1000 hours, standard deviation of 100 hours (assuming normal distribution)
  2. Mode B: Mean life of 1200 hours, standard deviation of 150 hours (also assuming normal distribution)

Using the Monte Carlo method, we would do this:

  1. Set the number of simulations: For instance, we simulate 10,000 times.
  2. For each simulation:

* Randomly draw a life value from Mode A's distribution (e.g., 950 hours).

* Randomly draw a life value from Mode B's distribution (e.g., 1100 hours).

* Since the product's life is determined by the "first mode to fail," the product life for this simulation is 950 hours (taking the minimum of the two).

  1. Repeat 10,000 times: This gives you 10,000 simulated product lives.
  2. Analyze the results: You can plot the distribution of these lives, calculate the mean life, standard deviation, and even find out the probability of the product failing within 500 hours.

You see, doesn't this allow for a more objective evaluation of your product's "expected lifespan"? That time, we used this method to simulate over ten failure modes, and the result showed that the probability of the product failing within one year was as high as 15%! The converted DPMO was approximately 150,000, which was shocking.

The Most Common Pitfall: Incorrect Data Source, Simulated Results are Garbage

Honestly, the most common pitfall with the Monte Carlo method is that your input data is fundamentally incorrect. If your initial estimation of each failure mode's life distribution and parameters (mean, standard deviation, or even distribution type, whether normal distribution or Weibull distribution) is inaccurate, then the simulated results are worthless.

I once saw a rookie directly applying results from laboratory "accelerated life testing" to products in real environments, and the parameters were completely off. The accelerated test life distribution is fundamentally different from the actual product degradation curve, yet he naively thought applying an Arrhenius equation would solve it. So, the key is that you must have reliable historical data or rigorous experimental designs to obtain these distribution parameters; otherwise, it's like the blind men describing an elephant.

One Thing You Can Do Today

Identify the two most critical failure modes for your product, estimate their life distributions, and then run 1000 Monte Carlo simulations in Excel.

Want to try it yourself?

Every tool mentioned in this article is available on InsightFab — just upload a CSV to analyze.

Go to Tools

More in this category

Reliability Fundamentals: The Relationship Between Failure Rate, MTBF, and Reliability Function

This article offers practical insights, starting with the author's experience of realizing reliability extends beyond just yield, focusing instead on a product's lifespan in customer hands. It uses relatable examples to clarify key reliability terms like failure rate, MTBF, and reliability function, helping engineers understand what truly matters to customers: stable, long-lasting product operation.

Weibull Analysis: Predicting Equipment Life from Failure Data

The Weibull distribution is the most commonly used life analysis tool in reliability engineering. Through the shape parameter β, it can be determined whether failures are early failures (infant mortality), random failures, or wear-out failures, thereby determining the correct maintenance strategy.

Reliability Prediction: Beyond MIL-HDBK-217, Reliability Estimation Engineers Can Do

Reliability prediction is not limited to consulting the MIL-HDBK-217 manual. This article explains reliability calculations for series and parallel systems, the benefits of derating design, the application of HALT testing, and how to avoid treating predicted values as guarantees.