Troubleshooting is a form of problem solving. It is the systematic search for the source of a problem so that it can be solved. Troubleshooting is often a process of elimination - eliminating potential causes of a problem. Troubleshooting is used in many fields such as system administration and electronics.
In general troubleshooting is the identification or diagnosis of "trouble" in a system. The problem or is initially described as symptoms of malfunction and troubleshooting is the process of determining the causes of these symptoms.
A system can be described in terms of its expected or intended behavior (usually, for artificial systems, its purpose). Events or inputs to the system are expected to generate specific results or outputs. (For example selecting the "print" option from various computer applications is intended to result in hardcopy emerging from some specific device). Any unexpected, particularly undesirable behavior is a symptom and troubleshooting is the process if isolating its specific cause or causes. Frequently the symptom is a failure to observe any results. (Nothing was printed, for example).
Most discussion of troubleshooting, and especially training in formal troubleshooting procedures, is extremely domain specific. The bulk of the material is relevant to a particular field of endeavor (such as automotive repair, computer hardware services, or software systems support). However, troubleshooting has common elements regardless of the specifics.
Any system can be described in terms of its components or subsystems. Each subsystem can be described in terms of its expected behavior. So the inputs to a system can be described as a cascade of inputs and results among the components of the system. (For example: selecting the "print" option in a computer application may cause the software to call on a separate utility, such as lpr on a UNIX system; that in turn might open, read and parse a number of configuration files which might direct it to perform some form of hostname address resolution via DNS, NIS, or LDAP, and then initiate a TCP/IP connection to a specific network device, and so on).
The domain specific knowlege that dominates the troubleshooting process is largely comprised of the understanding of these systems in terms of the interactions and dependencies among their subsystems and components. In particular the specialist can ennumerate the components and knows a set of procedures for testing many of them in isolation from the system as a whole. (For example the systems administrator may know which configuration files lpr is trying to parse and may read them manually, check their permissions, or may assume the identity of the user who is experiencing the problem and manually run an lpr command from the system's shell prompt; this may isolation the problem to the application's configuration, the user's preference settings, the workstation's configuration or network settings, the network's name services domain, or back to the printer's configuration or hardware).
Well-designed systems have designated "test points" or monitoring instrumentation. (For example most printers have indicator lights which change colors or blink, or LCD panels which display messages for detectable problems: paper jams, empty paper trays, network or other cable disconnection, etc. As another example UNIX and Linux systems support features for system call tracing through commands like truss, strace, and ktrace).
Usually troubleshooting is applied to something that has suddenly stopped working, since its previously working state forms the expectations about its continued behavior. So the initial focus is often on recent changes to the system or to the environment in which it exists. (For example a printer that "was working when it was plugged in over there"). However, there is a well known principle that correlation does not imply causality. (For example the failure of a device shortly after it's been plugged into a different outlet doesn't necessarily mean that the events were related. The failure could have been a matter of coincidence).
It's useful to consider the common experiences we have with light bulbs. Light bulbs "burn out" more or less at random; eventually the repeated heating and cooling of its filament, and fluctuations in the power supplied to it cause the filament to crack or vaporize. The same principle applies to most other electronic devices and similar principles apply to mechanical devices. Some failures are part of the normal wear-and-tear of components in a system.
A basic principle in troubleshooting is to start from the simplest and most probable possible problems first. This is illustrated by the old saying "When you see hoof prints, look for horses, not zebras", or to use another maxim, use the KISS principle. This principle results in the common complaint about help desks or manuals, that they sometimes first ask: "Is it plugged in and does that receptacle have power?", but this should not be taken as an affront, rather it should serve as a reminder or conditioning to always check the simple things first before calling for help.
A troubleshooter could check each component in a system one by one, substituting known good components for each potentially suspect one. However, this process of "serial substitution" can be considered degenerate when components are substituted without regards to a hypothesis concerning how their failure could result in the symptoms being diagnosed.
Efficient methodical troubleshooting starts with a clear understanding of the expected behavior of the system and the symptoms being observed. From there the troubleshooter forms hypotheses on potential causes, and devises (or perhaps references a standardized checklist) of tests to eliminate these prospective causes. Two common strategies used by troubleshooters are to check for frequently encountered or easily tested conditions first (for example, checking to ensure that a printer's light is on and that its cable is firmly seated at both ends), and to "bisect" the system (for example in a network printing system, checking to see if the job reached the server to determine whether a problem exists in the subsystems "towards" the user's end or "towards" the device).
This latter technique can be particular efficient in systems with long chains of serialized dependencies or interactions among its components. It's simply the application of a binary search across the range of dependences.
Simple and intermediate systems are characterized by lists or trees of dependencies among their components or subsystems. More complex systems contain cyclical dependencies or interactions (feedback loops). Such systems are less amenable to "bisection" troubleshooting techniques.
It also helps to start from a known good state, the best example being a computer reboot. A cognitive walkthrough is also a good thing to try. Comprehensive documentation produced by proficient technical writers is very helpful, especially if it provides a theory of operation for the subject device or system.
A common cause of problems is bad design, for example bad human factors design, where a device could be inserted backward or upside down due to the lack of an appropriate forcing function (behavior-shaping constraint), or a lack of error-tolerant design. This is especially bad if accompanied by habituation, where the user just doesn't notice the incorrect usage, for instance if two parts have different functions but share a common case so that it isn't apparent on a casual inspection which part is being used.
Troubleshooting can also take the form of a systematic checklist, troubleshooting procedure, flowchart or table that is made before a problem occurs. Developing troubleshooting procedures in advance allows sufficient thought about the steps to take in troubleshooting and organizing the troubleshooting into the most efficient troubleshooting process. Troubleshooting tables can be computerized to make them more efficient for users.

Reproducing Symptoms
Some of the most difficult troubleshooting issues related to symptoms which are only intermittent. In electronics this often is the result of components which are thermally sensitive (since resistence of a circuit varies with the temperature of the conductors in it). Compressed air can be used to cool specific spots on a circuit board and a heat gun can be used to raise the temperatures; thus troubleshooting of electronics systems frequently entails applying these tools in order to reproduce a problem. Another, extremely common, problem in electronic and electro-mechanical systems
In computer programming race conditions often lead to intermittent symptoms which are extremely difficult to reproduce; various techniques can be used to force the particular function or module to be called more rapidly then it would be in normal operation (analogous to "heating up" a component in a hardware circuit) while other techniques can be used to introduce greater delays in, or force syncronization among, other modules or interacting processes.
Steven Litt [1] defines intermittent issues thus:
In particular he asserts that there is a distinction between frequency of occurence and a "known procedure to consistently reproduce" an issue. For example knowing that an intermitten problem occurs "within" an hour of a particular stimulus or event ... but that sometimes it happens in five minutes and other times it takes almost an hour ... does not constitute a "known procedure" even if the stimulus does increase the frequency of observable exhibitions of the symptom.
Nevertheless, sometimes troubleshooters must resort to statistical methods ... and can only find procedures to increase the symptom's occurence to a point at which serial substitution or some other technique is feasible. In such cases, even when the symptom seems to disappear for significantly longer periods, there is a low confidence that the root cause has been found and that the problem is truly solved.