A system has two engines. When both engines are working, the failure rate of each engine is $1$ failure per $10$ days. When there is only one engine working, the failure rate is $1$ failure per $6$ days (since the single engine must work harder to compensate). Engines are repaired one at a time with a rate of $1$ repair per day. Assume all failure times and repair times are exponentially distributed and are also independent.

(a) Find the expected time to reach the failure state, starting from a point where both engines are working?

(b) If the system is currently failed (aka, both machines not working), find the probability that the system reaches the "both-working" state BEFORE returning to the failed state?

My attempt: (a) From the given information, the transition rate matrix is: $$Q =\pmatrix{\frac{-1}{5} & \frac{1}{5} & 0\\ 1 & \frac{-7}{6} & \frac{1}{6}\\ 0 & 2 & -2}$$ Thus the probability matrix is $$P =\pmatrix{0 & 1 & 0\\ \frac{6}{7} & 0 & \frac{1}{7}\\ 0 & 1 & 0}$$ Now, solving for $p_0$ - which is the fraction of time of being in the failed state - from the equation $PQ=0$, we get: $p_0 = \frac{1}{73}$. We also find $\pi_0 = \frac{3}{7}$ - the long-run probability of being in the failed state in the long run - from solving the equation: $\pi P =\pi$. Thus the expected time to reach the failure state is $p_0\pi_0 = \fbox{$\frac{3}{511}$}$.

(b) To move from the current failed state to "both-working" state BEFORE returning to the failed state means the system has to go from state 1 (one of the machine is working) to state 2 (both machine works), since the system moves from failed state to state 1 with probability $1$. Thus, it has to reach state 2 at the next step (otherwise, it would return to the failed state BEFORE reaching state 2), so the required probability is $\fbox{$\frac{6}{7}$}$.

My question: I wonder if my solution to part (a) is correct, since $\pi_0$ also accounts for probability of moving from state 1 (one of the machine is working) to the failure state, so it might be more than just $\fbox{$\lim_{n\rightarrow \infty} P^{(n)}_{20}$}$. I think my solution to part (b) is correct, but if I'm wrong, please help point out the reason why.


For problem (a):

Let $x$ be the expected time to double failure. The expected time to the state of one engine working is $5$ (everything here is in days). From the state of one engine working, the expected time to leave that state is $\frac67$, and when you leave, the probability that you have left into the total failure state is $\frac17$. If you leave into the no-failure state, you have to start again. So... $$ x = 5+\frac17\left(\frac67\right) + \frac67\left(\frac67 + x\right) $$ The solution is $x = \frac{291}{7}$ or just over $41\frac12$ days. I really don't see how you could think the expected time to a double failure was less than $\frac1{80}$ of a day.

For problem (b):

From the doubly failed state you will certainly next reach the state of one engine working. From there, as in part (a), there is a $\frac17$ chance of returning to the doubly failed state, and a $\frac67$ chance of transitioning to the both-working state. So the answer is $\frac67$. (Which is the answer you arrived at.)

You can actually obtain the eigenvalues and eigenvectors of the transition matrix analytically without solving a cubic equation since the constant term in the characteristic equation is zero. Using those eigenvectors as new variables, you can get the probability of each state as a function of time, in closed form. But the two questions you were asked have much easier solutions, as above.

  • $\begingroup$ Beautiful solution to part (a) (beautiful problem as well)! Thanks a lot for your detailed explanation. $\endgroup$ – user177196 Oct 23 '16 at 5:01
  • $\begingroup$ The result should be $41$ days though. I got $x = \frac{287}{7}=41$ $\endgroup$ – user177196 Oct 23 '16 at 5:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.