The　overall　performance　of　complex　industrial　systems　using　control　and　optimization　strategies　is　often　hampered　by　the　lack　of　available　measurements　of　key　quality　variables.　The　development　of　a　soft　sensor　is　a　solution　that　allows　for　the　necessary　process　information　to　be　made　available.　Due　to　the　nonlinear　characteristics　of　industrial　processes,　a　new　soft　sensor　development　approach　that　incorporates　infrequent,　variable　time　delayed　measurements　is　proposed.　Based　on　the　fast-sampled　measurement　and　the　nonlinear,　first　principles　model,　the　frequent　and　delay-free　key　performance　indicator　(KPI)　estimation　is　realized　using　the　square-root　unscented　Kalman　filter　(SR-UKF).　Then,　a　modified　Kalman　Filter　(MKF)　algorithm　is　proposed　to　deal　with　those　infrequent,　but　accurate　measurements,　whose　values　are　variably　delayed　in　time.　These　two　types　of　estimates　are　fused　and　optimized　to　give　an　optimal,　reliable　KPI　estimate　based　on　the　distributed　state　fusion　UKF　filter　algorithm.　The　performance　and　effectiveness　of　the　proposed　approach　are　demonstrated　by　an　experimental　application　for　alumina　concentration　monitoring　in　the　aluminum　electrolysis　industry.