how about using inferschema one single time to create a correct DF, then create a schema from the df-schema.

something like this f.e.

from pyspark.sql.types import StructType    
 
# Save schema from the original DataFrame into json:
schema_json = df.schema.json()
 
# Restore schema from json:
import json
new_schema = StructType.fromJson(json.loads(schema_json))