Recommendations for rationalizing cleaning-in-place in the dairy industry: Case study of an ultra-high temperature heat exchanger

No information is given about the launching and production phases because the industrial operators did not choose to change any operating parameters because the changes might have altered the final cream characteristics. Nonetheless, the differential pressure sensor gave relevant information about fouling evolution during the production phase.

The current management of the intermediate and final rinsing steps generated losses evaluated at approximately 1,500 L of water ( Table 1 ) containing 20 kg of caustic soda and 14 kg of nitric acid. Reducing the rinsing durations would reduce the volume of effluents, and the recovery of the detergent solution could also be realized. As for the flushing–first rinse phase, the effluent could be separated into 2 fractions by using the conductivity sensor signal; at the very beginning, a diluted detergent solution could be recovered. The lower the sorting threshold of caustic soda concentration, the higher the dilution and the volume of recovered detergent, the stronger the readjustment of detergent concentration, and the bigger the storage capacity required. The other fraction to be recovered (very diluted detergent solution) could be either disposed of and sent to the purification station or recovered as flushing–first rinse water, but in that case, it would be necessary to control the sorting of the cream from launching water so as to avoid any traces of detergent in the manufactured product.

During the intermediate rinses, the detergent concentration sharply decreased after 4.6 ± 0.3 min and 4.8 ± 0.5 min for NaOH and HNO, respectively ( Figures 4a and 6 ). It then leveled off, indicating that the flushing operation was over. Regardless of the operation, the critical detergent concentration of 0.1 g/kg was reached after less than 11 min ( Figure 4a ). For the final water rinse (results not shown), an acid concentration lower than 20 mg/kg (0.91 mS/cm) was reached after less than 9 min, indicating that the duration of the final rinsing phase could have been significantly reduced.

The HNO-2 phase was similar to the HNO-1 phase. The calcium concentration (maximum 500 mg/kg, compared with 4,500 mg/kg in the HNO-1 phase) became equal to that of the water from 5 to 6 min and α decreased rapidly to reach a stable level of 0.5 ± 0.3% ( Figure 4a Table 1 ). The total amount of calcium removed, around 17 g, was 16 times less than that removed during the HNO-1 phase. The current duration of 18 min could have been decreased to 6 to 7 min.

During the NaOH-2 phase, the average α level at the end of the cleaning (α = 7.4 ± 0.5%) was higher than after the NaOH-1 and HNO-1 phases ( Figure 4a Table 1 ). No simple explanation could be given with regard to residual fouling because the sterilizer could be considered to be clean at the end of the first alkaline and acid phases. The same trend was observed with water at the end of the alkaline rinses (intermediate rinses; Figure 4a ). There are then unexpected variations of α when there is a change of fluid in the sterilizer. One can suppose that the differential pressure sensor or the removal of the fouling layer could be sensitive to a change in operating conditions (e.g., temperature, flow-rate, and physical and chemical characteristics of the fluid) or a change in physical and chemical characteristics of the fouling materials extracted from the sterilizer surface. It was likely that, because of the effect of OH- ions or even water, the residual deposit hydrates and swells as shown previously during the NaOH-1 phase but without being totally removed during the NaOH-2 phase. These observations are in agreement with the results and discussion reported by, who observed an increase of the fouling rate of the heating zone of a plate UHT sterilizer fouled by milk from the very beginning to the end of the alkaline cleaning. Such a phenomenon was observed only when NaOH was the chemical reactant used for the first phase of cleaning (); in case of an alkaline cleaning operated after an acid cleaning, the phenomenon of swelling was not observed, contrary to our results.

During the NaOH-2 phase and regardless of the cleaning time, the maximum C COD value was very low (less than 40 mg/kg), with an average value of 10 mg/kg (results not shown). Because the removed fouling amount was approximately 250 times lower than that measured during the NaOH-1 phase, the necessity of the NaOH-2 phase was questionable. It was likely that the small existing fouling would not alter the microbiological safety of the end product because it probably stayed in the end of the heating section where the temperature was more than 100°C. Nonetheless, a risk is still possible, and the residual deposit could act as an agent for future fouling. If this phase has to be retained, its duration should be significantly reduced from 18 to 10 min as for the NaOH-1 phase.

During the HNO-1 phase, α decreased after approximately 3 min of cleaning (average α = 2.8 ± 0.5%; Figure 4a ), as reported bywith a UHT sterilizer fouled with milk. The fast cleaning rate was confirmed by the fast calcium removal, the concentration of which become equal to that of the water after 5 to 6 min (e.g., approximately 1 min after the flow of the first Hions; Figure 5 ). The overall amount of removed calcium was around 270 g. These results were in agreement with, who shows that the overall amount of calcium contained in the fouling of a tubular sterilizer of whole milk is removed after less than 5 min. Thus, the acid cleaning duration could have been sharply reduced from 18 min to 6 to 7 min.

During the NaOH-1 phase, a significant pollution load (C= 3–10 g/kg), not removed by the flushing–first rinse operation, was expulsed with water from the equipment ( Figure 3 ). However, for the first 4 min of the NaOH-2 and acid phases, the effluent was mainly clean water (results not shown); the first 4 min could therefore have been operated in a closed loop so as to save 450 L of water per phase, corresponding to a total amount of 1.35 mof water per cleaning. After 5 min, the flow of the chemical cleaning solution was detected through the sudden increase of Cor C(equations 3 and 4 ), followed by oscillations resulting from the sequential injection of the concentrated detergent in the feed tank ( Figures 3 and 4a ). The oscillations, the period of which was equal to the mean residence time in the closed loop (approximately 5 min), were attenuated by progressive mixing of the detergent in water contained in the closed loop. During the first 2 min of the NaOH-1 phase (4–6 min after the beginning of the phase), ΔP ( Figure 3 ) and α ( Figure 4a ) increased simultaneously with the increase of OHions in the heating zone of the sterilizer. Such an increase was a result of the swelling of the deposit in contact with caustic soda, as previously reported for a UHT plate sterilizer of chocolate dessert cream () and for various plants heat-treating dairy fluids (). Then, ΔP and α decreased progressively, showing that cleaning continued, and stabilized after approximately 10 min of cleaning ( Figures 3 and 4a ). The turbidity signal was difficult to use or interpret because it fluctuated with time ( Figure 3 ), even though it attenuated after 10 min. The cleaning did not seem to be efficient during the following 8 min (average α = 5.1 ± 0.3%; Figure 4a Figure 3 ) and could have been be shortened.

The chemical cleaning procedure consisted of 4 phases. At the beginning of each phase, the concentrated detergent was directly injected into the feed tank. The first 4 min, operated with an open loop, yielded approximately 450 L of effluents that was disposed of at the purification station ( Table 1 ).

The critical analysis of this phase showed 3 possible improvements. 1) The treatment of the conductivity enabled the sorting of the fluids for further valorization. For example, sorting the fluids at 13.6 min can lead to the separate collection of approximately 320 L of effluent with an average DM of 60 g/L and 130 L of effluent with an average DM of 29 g/L ( Figure 2b ). Moreover, knowing the acceptable water content of the end product, a small but not negligible quantity of little-diluted cream recovered during the very beginning of the phase could be reintroduced in the production line. 2) The DM information showed that it could be interesting to increase the flow-rate right at the beginning of this step. Moreover, the final part of the first rinse phase, operated in a closed loop, could be changed to an open loop because the beginning of the first alkaline cleaning phase that follows immediately was operated that way. To choose adequate parameters (flow-rate, running time, open or closed loop) for this phase (DM <1 g/kg), the balance between energy consumption and saved time should then be estimated on the basis of newly collected experimental results. 3) Instead of increasing the duration of this phase to reach the industrial goal of DM <1 g/kg, the effect of small amounts of matter on the cleaning efficiency of NaOH should be studied. For example, stopping the operation at 12 min (instead of 16 min) would lead to the addition of 13 kg of DM in the NaOH solution (DM collected from 12 min to the end of the water rinse phase; i.e., 16 min) and then to an increase of Cin the NaOH solution (from 7.4 g/kg to 22.4 g/kg in the studied case). Such added pollution may be not detrimental to the NaOH solution efficiency because the presence of fat and proteins solubilized in the caustic soda solution was previously shown to decrease surface tension characteristics of the solution, leading to interesting cleaning properties of the detergent solution ().

The flushing–first rinse phase was carried out in 2 cycles, both performed successively in open and closed loops, the second cycle being performed at a high flow-rate (6 instead of 2 m/h) ( Figure 2a Table 1 ). During this phase, DM sharply decreased during the first minutes and leveled off for 7.5 < time (min) < 12.3. After 12.3 min, the increase of Q and then of the shear stress at the sterilizer surface favored the soil removal and the increase in DM, as previously noted byand. A subsequent increase of waste DM was observed. The threshold at the end of this phase, set to 1 g/kg by the industrial operator, was not reached, suggesting that rinsing time (duration and waste volume) should have been increased. The matter losses during these 2 phases were evaluated at 22.5 ± 3.0 kg of DM in 450 ± 15 L of effluent ( Figure 2b Table 1 ).

In this study, alkaline cleaning was realized according to single-use CIP mode. During cleaning, NaOH is polluted with suspended solids and soluble compounds ( Figure 3 ). It would be of interest to study alkaline phase kinetics and duration by using reused NaOH treated by microfiltration (so as to eliminate suspended solids) to quantify the effect of NaOH solutions with lower surface tension as shown byfor UF membrane cleaning.

Despite the satisfactory industrial results given by the revised sequence, further improvements, listed as follows, might still be made. 1) For improving cleaning, the caustic concentration should be controlled carefully after having experimentally defined the optimal value. 2) The HNO-2 phase is useless and can be removed. 3) The order of cleaning sequences should be validated in terms of cleaning kinetics and efficiency. Indeed,reported that for a UHT sterilizer of skim milk, the cleaning duration of the preheating zone (80–120°C) was 2 times faster when the acid cleaning was operated before alkaline cleaning; moreover, only the order acid-base was efficient for cleaning the heating zone (120–140°C). The enhanced cleaning rate with the order acid-base was also demonstrated byfor cleaning a tubular evaporator (70°C) fouled with whey. 4) The relationship between sterilizer fouling level (related to production duration) and chemical cleaning rate and duration should be studied. It would help control the combination of production duration and cleaning duration for better UHT sterilizer management. 5) Supplementary savings of matter should be validated by a more adequate management of the sorting and use of fluids (e.g., methane bioproduction, external substrate of denitrification, and so on;). The experiment requires a new plant for recovering the fluids, which was not industrially possible in the present study.

Compared with the initial sequence, α at the end of the final rinsing was smaller in the revised sequence (α = 0.0 ± 0.2% compared with 1.5 ± 0.3%; Table 1 Figure 4 ); this was unexpected because the revised sequence included only 1 shortened alkaline cleaning. This difference was not the outcome of a different initial level of fouling (similar at the end of the first rinsing phase) or of different thermal and hydrodynamic cleaning and rinsing conditions, but originated from a gap in detergent concentrations. These concentrations were much smaller for the revised sequence owing to large concentration variations of mother solutions injected. For NaOH, the concentration was in the range of 0 to 30 g/kg versus 0 to 50 g/kg for the initial sequence; for HNO, the concentration was in the range of 0 to 25 g/kg versus 0 to 45 g/kg for the initial sequence. Such differences could generate significant differences of efficiency. According to, the cleaning duration of a deposit of whole milk (72°C) is multiplied by approximately 2 when alkaline detergent increases from 30 to 50 g/kg. According to, the cleaning duration of stainless steel tubes fouled with whey (97°C) is multiplied by 5 when the caustic concentration increases from 5 to 20 g/kg. This trend was also observed after the alkaline rinsing: the average caustic concentration was close to 20 g/kg in the revised sequence compared with 34 g/kg in the initial sequence and this leads to a fouling rate (α) 2 times smaller with the revised sequence. This leads to the idea that the removal of a larger fraction of fouling by the alkaline cleaning at lower concentration made easier the access of Hions to the wall, where the deposit, mainly inorganic, was eliminated by the acid action. This kind of reasoning, based upon the average concentration, applied to the highly fluctuating values of concentration within the sterilizer.

With the revised cleaning sequence (operated twice for the current study at industrial level), the α at the end of the final rinsing (α = 0.0 ± 0.2%) was equal to or lower than initial α determined during the launching phase (2.8–5.4 ± 0.6%) and much lower than α reached after the production phase. These low α values demonstrated the high efficiency of the revised sequence. The NaOH-1 phase removed the major part of fouling (0.5 ≤ the fouling fraction removed by a phase ≤ 0.9), and the residual fouling fraction was removed with the subsequent acid cleanings (HNO-1 and HNO-2 phases). When the NaOH-2 phase was not operated, the HNO-2 phase removed no fouling fraction, confirming the usefulness of that phase ( Table 1 ).

The gains induced by means of the industrially revised sequence were considerable in duration and overall effluents volume, both divided by 2 ( Figures 4a and 4b Table 1 ). The losses of NaOH and HNOwere reduced with the revised sequence: 11 ± 3 compared with 43 ± 4 kg of DM and 16 ± 3 compared with 28 ± 1 kg of DM for NaOH and HNO, respectively.

The new parameters of the revised sequences to be checked were as follows ( Table 1 ). For the flushing– water rinse phase, the overall duration of the phase remained unchanged and the last step, originally operated in a closed loop, was operated in an open loop. The duration of the NaOH-1 phase was reduced from 18.2 ± 0.2 to 16.1 ± 0.2 min. For both the HNO-1 and HNO-2 phases, the initial duration of 18.2 ± 0.2 min was reduced to 7.2 ± 0.5 min and the first 4 min were operated in a closed loop. The NaOH-2 phase was suppressed with the suppression of the following intermediate rinsing. Because the NaOH-2 phase was suppressed, the HNO-2 phase was likely useless, but the industrial partner decided to retain it. Regarding intermediate and final rinses, the automatic program was adapted to control the end of operation on-line with the conductivity sensor. For safety reasons, a calculated delay was inserted for controlling the minimum and maximum duration of the operation so as to mitigate any dysfunction of the sensor. The durations of the intermediate rinsing steps after NaOH and HNOcleanings were reduced from 12.5 ± 0.3 min and 16.8 ± 0.7 min, respectively, to 7.0 ± 0.3 min for each. The duration of the final rinsing was reduced from 14.0 ± 0.2 min to 7.8 ± 0.2 min. The conductivity and pH at the outlet of the plant were consistent with those of drinking water (χ ≤ 1,000 μS/cm; 4.5 < pH ≤ 9.0).

The preceding critical analysis of the initial industrial cleaning sequences made it possible to propose completely revised sequences. Among all the modifications suggested, many of them were accepted by the industrial operators. Their relevance was then quantified and checked on an industrial scale. The operating conditions of each phase and the volume of the effluents disposed of to the purification station during both sequences are reported in Table 1

Strategy for Rationalizing Cleaning

This work opens the way to a general procedure for the rational management of CIP, which requires 4 main steps. To propose improvements, the first compulsory step consists of knowing the general scheme of the plant and the geometrical process characteristics from the CIP tanks to the CIP backflow and the waste valves. This knowledge eventually allows the fixing of equipment design mistakes that have created dead-zones sources resulting in rinsing and cleaning difficulties, and the determination of the time scales to be synchronized and the volumes involved in the plant.

In the second step, the appropriate sensors or tracers and their signal treatment methods for each phase of the process must be selected. Two main sensors should be exploited: the turbidity sensor, which allows the end of water flushing–first rinse phases to be determined and could also be useful for operating the sorting of fluids during the launching phase, and the conductivity sensor, which determines the time of the end of water flushing–first rinsing phases as well as the rinsing phases. It could also be appropriate for sorting both the cream during the launching and first rinse phases and the cleaning solutions (NaOH, HNO3) during the rinses. Both sensors are important for knowing the nature and the concentration of each fluid flowing in the equipment in real time, the knowledge of which opens the way to a better control of the process. Special attention should be paid to the treatment of the signal of the pressure-difference sensor, which quantifies the fouling rate throughout the process. It quantifies the state of the sterilizer and the cleaning efficiency at any processing time, in particular at the beginning and the end of each phase and of the entire sequence. The off-line analyses of both CCOD and calcium concentrations are also useful for evaluating the amount of fouling matter removed from the food plant surface by the alkaline and acid cleanings, respectively.

In the third step, the database must be designed. This allows all the various data collected from the sensors or tracers during the different phases (launching, production, cleaning, rinsing) to be acquired. The adequate selection of the acquisition frequency must be adapted to each phase and to each step of the phase according to its duration and to the kinetics of concern.

The final step consists of exploiting the data according to the calculation modes selected for synchronizing the time scales; calculating the parameters useful for the evaluation of fouling, rinsing, and cleaning kinetics; and determining the operation ends and the sorting of the fluids mixtures and cumulated volume, effluents load. Such data exploitation allows an on-line management to be implemented that performs better than a temporization. Also, the data is the source of ways in which the production and cleaning modes of the equipment can be improved.