More than one operator should carry out the inspection and cleaning of equipments in a cleaning plant because of the risk of heatstroke. However, due to the shortage of operators, most inspections are often done by only a single operator. Moreover, clean the equipments alone. Therefore, an indoor positioning system to track the location of operators is needed to detect accidents as soon as possible. This study focuses on magnetic fingerprinting, which does not require infrastructure deployment, as a method for the indoor positioning system in a cleaning plant. Magnetic singularity is observed at different points because many magnetized metals, large motors, and generators are in the cleaning plant. This study aims to develop an indoor positioning system using magnetic fingerprinting for a cleaning plant, and magnetic data is collected from an actual cleaning plant for analysis. This paper proposes to separate the vertical and horizontal components of the spatial magnetic field by scalar projection and treat them as magnetic fingerprints to capture the magnetic features in more detail to improve positioning accuracy. We evaluate the positioning accuracy of a magnetic fingerprinting path matching method that treats neighboring magnetic fingerprints as a combined time series data based on this magnetic fingerprint. In the evaluation, we experimented with calculating the positioning error using the data obtained from a cleaning plant. This experiment is based on the hypothesis that the positioning performance of the proposed method using the magnetic fingerprint is higher than that of the existing method using the composite scalar of the spatial magnetic field as the fingerprint. We evaluated the performance of the proposed method assuming an indoor positioning system in a cleaning plant. The results show that the proposed method can be applied to the indoor positioning system in a cleaning plant. As a result, in the verification experiment where data was collected using a measurement cart that allowed the position, height, and orientation of the positioning terminal to be fixed, the fingerprints of the same position were acquired in both the offline and estimation phases, the average positioning error of the existing method was 0.07~m, while that of the proposed method was 0.05~m This result is consistent with the hypothesis.