Indoor Positional Navigation (IPN) systems can be used to track the position of tools in factories which is crucial for quality assurance in many manufacturing industries. Inertial navigation is rarely used on its own because of the noisy Inertial Measurement Unit (IMU) sensors which contribute to large drift. Current IPN systems usually involve the installation and calibration of cameras or antennas, so achieving sufficient accuracy with inertial navigation based IPN would be very desirable. This project aims to evaluate an inertial navigation algorithm, based on Zero Velocity Update (ZUPT), for bolt level positioning by repeatability tests using an industrial robot. The ZUPT algorithm, developed at Atlas Copco, manages to effectively reduce drift and achieve moderate accuracy in position for simpler movements. The gravity tracking Kalman filter dictates the systematic errors in position that grow large with increased degree and dimension of rotation. When keeping rotations within 45◦ for a linear movement the absolute error in position is under 10%. Frequent stops are important when moving in a more complex trajectory to be able to negate drift, consequently detecting the start and stop of motion is crucial. The results show that increased frequency will improve accuracy. It is shown that averaging IMU samples before calculations can increase both truthfulness and precision by 10−25%, if sampling the IMU faster than the calculations. The ZUPT approach of inertial navigation will never yield positional results in real time, and the evaluated algorithm only performs well within certain limitations, mainly frequent stops and simple movements. Despite these limitations there is potential in using the algorithm for quality assurance purposes in hand held industrial fasteners. / Kvalitetssäkring är en central fråga för många tillverkningsindustrier, så som flygplans- och bilindustrin, där det är avgörande att varje förband har dragits åt på rätt sätt för att garantera säkerheten i produkten. Moderna fabriker har centrala styrsystem som kommunicerar med maskiner och verktyg, och ifall något blir fel är det vanligt att fabrikslinan stannar vilket blir kostsamt. Inomhuspositionering (IPS) av hög noggrannhet kan spåra vilken åtdragning som blivit fel, vilket dokumenteras och åtgärdas om möjligt senare, utan att stanna fabrikslinan. Dagens noggranna IPS system för kvalitetssäkring kräver installation och kalibrering av kameror och/eller antenner. Tröghetsnavigering kräver i grunden bara billiga sensorer installerade på verktyget men metoden är mycket opålitlig på grund av sensorernas opålitlighet och brus. I detta projekt har en metod för tröghetsnavigering, användandes Zero Velocity Update (ZUPT), evaluerats för kvalitetssäkring av handhållna verktyg genom repetabilitetstester. Tröghetsnavigeringsalgoritmen som tidigare utvecklats på Atlas Copco lyckas på effektivt sätt reducera drift och uppnår rimlig noggranhet för enklare rörelser. För linjära rörelser med rotationer under 45◦ så erhålls ett absolut positionsfel inom 10%. För att fungera väl även för mer komplexa rörelser krävs frekventa stop, och noggrann rörelsedetektion är central. Denna ZUPT-metod kommer aldrig att kunna generera position i realtid och algoritmen presterar väl endast inom vissa begränsningar. Trots detta så finns god potential för metoden inom kvalitetssäkring för handhållna industriverktyg.

Inertial wearable sensors constitute a booming industry. They are self contained, low powered and highly miniaturized. They allow for remote or self monitoring of health-related parameters. When used to obtain 3-D position, velocity and orientation information, research has shown that it is possible to draw conclusion about issues such as fall risk, Parkinson disease and gait assessment. A key issues in extracting information from accelerometers and gyroscopes is the fusion of their noisy data to allow accurate assessment of the disease. This, so far, is an unsolved problem. Typically, a Kalman filter or its nonlinear, non-Gaussian version are implemented for estimating attitude â?? which in turn is critical for position estimation. However, sampling rates and large state vectors required make them unacceptable for the limited-capacity batteries of low-cost wearable sensors. The low-computation cost complementary filter has recently been re-emerging as the algorithm for attitude estimation. We employ it with a heuristic drift elimination method that is shown to remove, almost entirely, the drift caused by the gyroscope and hence generate a fairly accurate attitude and drift-eliminated position estimate. Inertial sensor data is obtained from the 10-axis SP-10C sensor, attached to a wearable insole that is inserted in the shoe. Data is obtained from walking in a structured indoor environment in Votey Hall.