1、 Daily maintenance

Sensor cleaning

Regularly inspect the surface of the sensor probe. If dust or oil stains are found, use a specialized cleaning agent to gently wipe to avoid damaging the optical components. For contact probes, it is necessary to ensure that there is no wear or corrosion on the contact surface with the material; For non-contact probes, it is necessary to check whether the installation position is offset to prevent signal attenuation. After cleaning, it is necessary to blow away residual liquid with dry air to avoid moisture interference.

Cable and connection inspection

Check the connections of sensors, instrument control boxes, and power cables weekly for looseness, with a focus on identifying signal interruptions or noise caused by poor contact. If the outer sheath of the cable is found to be damaged, it should be replaced immediately and insulated to prevent the risk of electrical leakage.

environmental control

The equipment should be deployed in a dry and ventilated environment, with temperature controlled at 15-30 ℃ and humidity<80% RH. Avoid direct sunlight or strong electromagnetic interference sources. If the environmental temperature and humidity fluctuate greatly, a temperature and humidity monitoring module should be added and linked to the air conditioning system.

Optical system maintenance

Monthly performance testing is conducted on the light source (halogen lamp/LED) and detector (InGaAs/PbS) to verify spectral intensity stability through standard samples. If the light source attenuation exceeds 20% or the detector noise increases, the components should be replaced in a timely manner.

2、 Model validation

Calibration of standard samples

Use standard samples covering the measurement range (high, medium, and low concentrations) for model validation every quarter to ensure that the predicted values deviate from the reference values by less than 5%. If the deviation exceeds the limit, it is necessary to re collect the sample spectrum and update the PLS model.

Dynamic environment verification

In the production process, 10% of the samples need to be taken from each batch of drying process for laboratory analysis (such as Karl Fischer method), comparing online predicted values with offline true values, and evaluating the adaptability of the model when the material matrix changes.

Abnormal data warning

Real time monitoring of spectral similarity (cosine angle method>0.99) and predicted residuals (RSEP<1%). If there are consecutive outliers, the model recalibration process needs to be triggered to prevent misjudgment of the drying endpoint.