Investigating the effects of oxygen enrichment with modified zeolites on the performance and emissions of a diesel engine through experimental and ANN approach

•Oxygen purification was made using acid and base modified zeolites.•Feeding the purified air to the engine decreased CO, HC, smoke density, and BSFC.•Despite oxygen enrichment, the increase in NOX emission was limited.•The engine parameters were predicted with satisfactory error rates. Oxygen enrichment of intake air in internal combustion engines (ICE) is one of the effective methods used to reduce exhaust emissions and increase performance. In this study, oxygen enrichment process was performed with pressure swing adsorption (PSA) method using two zeolite types including acid-modified zeolite (AMZ) and base-modified zeolite (BMZ). In the experiments, clinoptilolite-type natural zeolite in 1.6–3 mm fractions was used. HCI, NH4NO3 and NaOH solutions were used to obtain AMZ and BMZ. Engine experiments were carried out using a four-stroke, single cylinder, direct injection, air-cooled diesel engine. In addition, by designing an artificial neural network (ANN) model with the capability to estimate the experimental results, its performance was tested. When the oxygen rate (21%) in the medium air, where the experiments were carried out, was compared with the oxygen rates in the air obtained with the use of AMZ and BMZ in zeolite filter (ZF) system, an increase of 17.14% and 21.90% was observed, respectively. As a result of the addition of oxygen-enriched air to the fresh air drawn into the combustion chamber in the ZF system, the carbon monoxide (CO), hydrocarbon (HC), smoke density and brake specific fuel consumption (BSFC) values decreased and the nitrogen oxide (NOX), vibration, noise, exhaust gas temperature (EGT) and brake thermal efficiency (BTE) values increased when compared to standard diesel engine (SDE). According to the obtained results, ANN estimated the exhaust emission and performance parameters with the regression values (R) of 0.99423–0.99990 and low error rates in the range of 0.002–6.93%.