The present invention relates to the operation of membrane filtration plants and more particularly to the regulating of such plants by predictive modelling of the clogging, for example by neural networks.
It is known that the use of membranes, especially ultrafiltration membranes, has become widespread in recent years, especially in the field of the production of potable or industrial water. The hollow-fibre membranes thus used allow the water quality requirements to be met, even should the resources be degraded.
At the present time, there is considerable research with the objective of improving the productivity of plants for producing potable or industrial water using such membranes. This research is based on knowledge of the various factors and phenomena involved in the filtration of surface water or other fluids of variable quality. The first factor limiting production by the membranes results from the deposition of particles on the surface and/or in the pores of the membranes. This first factor is a short-term phenomenon. To remove these particles, which are deposited on the membranes in the form of a layer or cake, hydraulic, pneumatic or hydropneumatic washing operations are periodically carried out. The second limiting factor is the adsorption of organic matter on the surface of the membranes and in the pores of the latter, this factor constituting a long-term phenomenon.
That part of membrane clogging that can be removed by hydraulic, pneumatic or hydropneumatic washing is often called reversible clogging, whereas the other part is called irreversible clogging.
There are many parameters involved in the clogging of the membranes used in water treatment. On the one hand, there are parameters relating to the quality of the fluid to be treated and, on the other hand, operating parameters, these two types of parameters being interdependent.
It will be understood that one of the ways of knowing how to increase the productivity of the filtration plant lies in having a better understanding of the phenomena involved in membrane clogging. For this purpose, one is led to modelling the membrane plant. Although a very large number of studies devoted to clogging have been carried out, the models produced are not applicable for describing the clogging of the membranes by complex fluids such as natural water. However, a number of promising tools allowing simulation models to be developed do exist. Among them, mention may be made of artificial neural networks. Such networks have been used successfully in predicting short-term performance. Moreover, it has been envisaged to develop a model for predicting the productivity of a plant for obtaining potable water, this prediction relying both on the quality of the water to be treated and on long-term operating parameters, taking into account the minimum number of parameters. In this regard, the reader may refer to the publication “Neural networks for long term prediction of fouling and backwash efficiency in ultrafiltration for drinking water production” by N. Delgrange-Vincent et al., published in Desalination 131, pp. 353-362, 2000.
Referring now to
This figure shows schematically an ultrafiltration module of the hollow-fibre type. The water to be treated is prefiltered beforehand and then injected using a pump P1 into the circulation loop of the module, a pump P2 circulating it in the loop.
The factors relating to the quality of the water are the following:
The plant operating parameters are the following:
The plant produces a constant permeate flow rate Qp, causing the pressure to rise during the filtration period. The circulation flow rate QC represents the feed rate at the inlet of the module. The membranes periodically undergo hydraulic washing with filtered water to which chlorine has been added. In this way, the level of membrane clogging is reduced.
The total hydraulic resistance of the ultrafiltration module is expressed by the equation:
R=Ptm/(μ.Qp/A)
where μ is the temperature-dependent viscosity of the water, Ptm is the average transmembrane pressure and A is the membrane area.
The total resistance is made up of the resistance of the membrane, the resistance due to reversible clogging and the resistance due to irreversible clogging. In the case of a constant permeate flow rate, the resistance builds up during the filtration period and decreases after backwashing, as shown in
Consequently, a production curve consists of cycles, each of them being characterized by the resistance (Re) at the end of the filtration cycle and the resistance (Rs) at the start of the next cycle, that is to say after hydraulic washing. Variations in the durations of the (Re) and (Rs) cycles therefore suffice to characterize and describe the variations in the filtration process.
The performance of a pilot production plant may be expressed through:
In the case of backwashing, the net flow rate is expressed by the equation:
Qpnet=(VF−VBW)/(tF+tBW)
in which:
The object of the present invention is to provide a method of regulating a membrane filtration plant designed so as to prevent irreversible clogging of the membranes while maximizing the productivity (estimated by a suitable criterion, such as the net production), whatever the quality of the fluid entering the system. In other words, the problem that has to be solved by the present invention consists in slaving the performance of a filtration plant to the quality of the incoming fluid; this slaving depends directly on the change in the clogging of the said plant, which change is predicted by neural network modelling so as to simulate the long-term operation of the filtration plant, the model allowing the plant to be monitored and controlled in real time.
If we consider the concept of the critical flux, as explained in the literature, it is preferable to operate with a flux low enough to completely avoid reversible clogging. Moreover, it has been observed that when the hydraulic resistance of the membranes increases at the start of a cycle, the amount of irreversible clogging increases with time. This observation means that the more the membrane is clogged, the greater the amount of irreversible clogging. A problem then arises which is due to the fact that the flux produced is extremely low when the treated water is of poor quality. A compromise consists in finding, for each cycle, the operating conditions such that, even if clogging does occur, it is possible to eliminate it by hydraulic washing and to ensure that this clogging is not irreversible.
To effect this regulation, it is possible to vary a number of operating parameters, which, as mentioned above, may be chosen from:
The present invention has adopted, as an example, for this regulation, on the one hand the filtration time and on the other hand the permeate flow rate, it being understood that other combinations of operating parameters may also be used without thereby departing from the scope of the invention.
It would be conceivable to work with a minimum permeate flow rate and a minimum filtration time so as to choose the most prudent approach with respect to the clogging phenomenon, but in this case the productivity would be too low. According to the invention, the productivity parameters, such as for example the permeate flow rate and the filtration time, are therefore varied so as to find a compromise between the highest water production on the one hand and the amount of clogging on the other, this compromise being quantified using a neural network model which calculates, according to the quality of the fluid to be treated and the state of the membrane for a given cycle, the change in the membrane permeability as a function of time, over a defined horizon, the quality of the fluid being simulated (constant or variable) over this horizon.
A priori, two situations may arise:
It was mentioned above that the state of the membrane at a given cycle may be characterized by its permeability, its hydraulic resistance at the start of a cycle or its transmembrane pressure. The method of regulation forming the subject-matter of the invention sets a clogging level limit at the start of the cycle, characterized by a permeability limit (Lp_c) and ensures that the plant operates with a permeability equal to or greater than this value.
Thus, according to the invention, at each cycle k the pilot plant will:
Two cases may be considered:
There remains to be defined what parameters have to be chosen in order to apply this regulation. It is necessary to choose the following:
This choice of regulating parameters is made using pilot plant regulating simulations.
These simulations were carried out according to the abovementioned strategy. To test the response of the model, six manipulations were made, during which the hydraulic resistance of the module was or was not made to drift. The corresponding water quality curves were plotted as a function of time.
At each cycle k, the experimental parameters and the operating conditions for the start of a cycle were introduced as input into the model and the neural network calculated, in loop mode, the hydraulic resistance over a horizon of H cycles starting from the assumption that all the input parameters are constant over these cycles. The permeability Lp_i(k+H) after H cycles was thus obtained and the net flow rate Qp_net_i was calculated.
All the (Qp;tF) pairs that could be applied to the next cycle were tested and, for each of them, the permeability Lp(k+H) after H cycles was calculated:
Next, the neural network is used to simulate the actual response of the pilot plant to the next cycle k+1, by inputting into it the permeate flow rate Qp and filtration time tF commands calculated beforehand, together with the new water quality and operating condition parameters. The network calculates the resistance at the end of the cycle and at the start of the next cycle.
To take into account possible large variations in the quality of the fluids to be treated, it is necessary to choose a horizon long enough to account for any drift in hydraulic resistance but, however, short enough for it to be possible to consider that the water quality is constant over the horizon H.
The permeate flow rate and filtration time limits and variation steps that have to be chosen in order to apply the regulation were also defined. The variation steps are the steps between the various flow rate and time values tested in order to optimize the net flow rate.
Finally, the influence of the choice of permeability limit value Lp_c on the controls and on the permeability drift was tested.
These simulations were used to validate the method of regulation of the invention using the neural network model to simulate the response of the pilot plant. It was thus possible to verify that the permeability was maintained at a particularly high level and that the net flow rate was high compared with a conventional operation without regulation.
This technique was then validated directly on site, on the pilot ultrafiltration plant.
The regulation algorithm was constructed. The essential points of the strategy on the basis of which this algorithm was constructed were the following:
The flowchart of the algorithm is illustrated by
The constants involved in the algorithm are:
The local variables are:
The call variables are:
In the “initializations” block, Qp_c and tF
The method of regulation forming the subject-matter of the invention was validated on site. An example of the results obtained over about one week of manipulation is illustrated by the curves in
Thanks to the invention, it has been possible to maintain a permeability above a fixed limit, for several days, by varying the filtration time tF and the permeate flow rate Qp in order to limit the amount of clogging of the ultrafiltration membranes.
Of course, it remains to be stated that the present invention is not limited to the embodiments described and illustrated above, rather it encompasses all variants thereof, such as those employing hydropneumatic or pneumatic washing operations or making use of operating parameters other than the permeate flow rate or the filtration time.
Number | Date | Country | Kind |
---|---|---|---|
00/16249 | Dec 2000 | FR | national |
Filing Document | Filing Date | Country | Kind | 371c Date |
---|---|---|---|---|
PCT/FR01/03828 | 12/4/2001 | WO | 12/1/2003 |