High temperature thermal storage systems are a key enabler of energy transition. China has over 50 MW-scale projects, with an average annual capacity expansion of 25%. However, operating temperatures of 300-1000°C are prone to failures such as molten salt freezing, pipe-line corrosion, and leakage, resulting in a 15%-20% unplanned system downtime rate and annual economic losses exceeding 10 million Yuan for a single 100 MW-scale system. Existing diagnostic methods suffer from bottlenecks such as long modelling cycles, strong sample dependency, and slow deployment. The low code platform cuts deployment from six months to 45 days. A 10 MW ANSYS FLUENT platform (4M grids) tested six scenarios (three freezing, the leakage, and ten reps each), proving the method's superiority for efficient operation. A fault characteristic model is constructed, including the molten salt freezing resistance coefficient and corrosion rate formula. Wavelet denoising and spatiotemporal interpolation are used to preprocess data. An algorithm is developed that integrates an improved RBF neural network (12 nodes in the input layer and 18 nodes in the hidden layer) with a fault knowledge base. A 10 MW system simulation platform (4 million structured grids) was built using ANSYS FLUENT. Six fault scenarios (three types of freezing and three types of leakage) were designed, with each scenario tested ten times. Results showed that this method achieved a diagnostic accuracy of 97.2%, an improvement of 18.7 and 7.9 percentage points, respectively, compared to the traditional threshold method (78.5%) and a single BP network (89.3%). The identification rate for minor faults was 92.3%, with a false alarm rate of 2.8%. The average diagnostic time was 1.8 seconds, supporting efficient system operation and maintenance.