Indirekt skattning av externomgivningstemperatur med internasensorer

Sammanfattning

This thesis investigates the feasibility of indirectly estimating the external ambienttemperature in a sealed camera system using only internal temperature sensors.The problem is particularly relevant for systems where external sensors cannot beused due to design or safety constraints, such as explosion-protected cameras.Experimental data was collected in a climate chamber over a temperaturerange of −40◦C to 50◦C, where internal sensor data and external reference temperature were recorded under controlled conditions. Based on this data, a datadriven model using multiple linear regression was developed, along with a hybridmodel that incorporates power consumption as a physical parameter.The results show that the data-driven model achieves high accuracy under stable operating conditions, with a mean absolute error (MAE) of 0.26◦C. However,the performance degrades when internal heat conditions change, such as whenthe heater is turned off. The hybrid model demonstrates significantly improved robustness, reducing the error to 0.43◦C in these scenarios by explicitly accountingfor internal heat generation.Furthermore, the results indicate that model performance is largely unaffectedby changes in orientation, suggesting that heat transfer is dominated by internalheat sources rather than convective effects.In conclusion, the study demonstrates that accurate estimation of ambienttemperature is possible using internal sensors alone. It also shows that combiningdata-driven methods with physical modeling is essential to achieve robustness andgeneralization across varying operating conditions.