NASHUA, N.H. — A modern airliner spends every flight taking measurements. Sensors determine how fast the aircraft is moving, how high it is flying and how its nose meets the airflow. Other systems track engine conditions and the movement of the aircraft itself.

Flight computers use that information to make calculations, operate automated systems and tell pilots what the airplane is doing. The challenge is that a sensor doesn't necessarily stop sending data when something goes wrong. It can continue producing numbers that look legitimate while describing something that isn't actually happening.

That leaves the aircraft with a problem beyond simply collecting accurate measurements. It also has to determine which information it can trust.

Related: Why turbulence is still difficult to predict

Redundancy gives the aircraft something to compare

One of aviation's basic answers is redundancy. Instead of depending on a single source for critical information, commercial aircraft can use multiple sensors or systems to measure the same parameter.

When those independent sources agree, the aircraft has more reason to trust the result. A substantial disagreement, however, gives onboard computers a reason to question the information and determine which source may be unreliable.

Imagine three sensors measuring the same condition. If two report similar values while the third suddenly moves far away from them, fault-detection logic can identify the outlier and continue using the remaining information. Aircraft systems have long used this kind of redundancy and comparison to maintain reliable data when individual sensors fail.

The problem gets more complicated when multiple sensors experience the same issue. Agreement doesn't automatically make a measurement correct, so engineers can't depend on a simple majority vote in every situation.

Computers can ask whether the data makes sense

Aircraft can also compare a sensor reading with what should be happening at that point in the flight.

An angle-of-attack (AOA) sensor offers a good example. The device measures the angle between the aircraft and the airflow around it, providing information that can feed other calculations and flight-control functions. Its readings should also change in predictable ways as the aircraft moves through different phases of flight.

Boeing's enhanced AOA system for the 737 MAX uses software monitors to check for some of those expected behaviors. During takeoff, for example, the AOA vanes should respond as the vehicle pitches up. Their behavior should change again as the airplane accelerates and aerodynamic forces increase.

If a sensor doesn't respond as expected, the monitoring software can identify its information as faulty. The system can then keep that data from reaching other systems or substitute a default value where appropriate. That allows the computer to consider more than whether two sensors agree. It can also determine whether a reading makes sense alongside everything else the airplane is doing.

Bad data can travel beyond the original sensor

Finding faulty data matters because modern aircraft systems share information. A measurement that starts at one sensor may feed several computers, which can use it to calculate other values or support automated functions.

For instance, an erroneous AOA measurement can affect calculated airspeed and altitude information as well as systems farther downstream. Boeing designed its enhanced AOA architecture in part to stop faulty information before it propagates through the aircraft.

Engineers therefore have to consider both detection and containment. Once a system recognizes questionable information, it needs to isolate that data while preserving enough reliable information for other functions to continue operating normally.

The same principle applies beyond AOA. As aircraft systems become more interconnected, knowing where a measurement came from and whether other systems can safely use it becomes part of managing a sensor failure.

The aircraft may have other ways to fill the gap

Engineers are also exploring ways to create redundancy without simply installing another physical sensor. Known as analytical or virtual redundancy, the approach uses other available aircraft information to estimate a value when the sensor that normally measures it becomes unreliable.

The concept takes advantage of relationships that already exist among different measurements. Airspeed, attitude and acceleration, for example, describe different parts of the aircraft's motion but don't behave independently. A computer model can compare those relationships and identify a measurement that no longer fits the larger picture.

NASA has studied virtual redundancy as a way to maintain usable information after sensor failures. Even if the device responsible for directly measuring a parameter stops providing trustworthy data, the vehicle may still have enough information elsewhere to estimate that value.

That doesn't eliminate the need for physical sensors. Instead, it gives the system another source of evidence when a direct measurement becomes questionable and another potential way to keep operating when that measurement is no longer available.

Sensors don't have to be perfect forever

Aircraft sensors work in an environment where temperature changes, moisture, ice, vibration and physical damage can affect equipment over time. Electronics and wiring introduce their own potential failure points, which means designers can't assume every measurement will remain accurate for the life of an airplane.

Instead, engineers design around the possibility that a sensor will eventually provide information the aircraft can't trust. The goal is to give the aircraft enough independent information to recognize when something doesn't add up and make sure one bad source doesn't get the final say.